reponame
stringclasses
5 values
filepath
stringlengths
15
64
content
stringlengths
63
24.6k
deep-diver/Continuous-Adaptation-for-Machine-Learning-System-to-Data-Changes
custom_components/batch_pred_evaluator.py
""" This component evaluates the performance of a currently deployed model, and the evaluation is based on the result of batch prediction on Vertex AI from the previous component. At the end, this component will output true or false to indicate if retraining is needed. Reference: https://bit.ly/vertex-batch """ from tfx.dsl.component.experimental.annotations import Parameter, OutputArtifact from tfx.dsl.component.experimental.decorators import component from tfx.types.experimental.simple_artifacts import Dataset from absl import logging import os import json @component def PerformanceEvaluator( gcs_destination: Parameter[str], local_directory: Parameter[str], threshold: Parameter[float], trigger_pipeline: OutputArtifact[Dataset], ): """ gcs_destination: GCS location where the files containing the result of batch prediction is local_directory: Temporary directory to hold files copied from the gcs_destination threshold: threshold to decide if retraining is needed or not it is based on the measured accuracy trigger_pipeline: an output artifact which hold true or false to indicate if retraining is needed or not """ full_gcs_results_dir = f"{gcs_destination}/{local_directory}" # Create missing directories. os.makedirs(local_directory, exist_ok=True) # Get the Cloud Storage paths for each result. os.system(f"gsutil -m cp -r {full_gcs_results_dir} {local_directory}") # Get most recently modified directory. latest_directory = max( [os.path.join(local_directory, d) for d in os.listdir(local_directory)], key=os.path.getmtime, ) # Get downloaded results in directory. results_files = [] for dirpath, subdirs, files in os.walk(latest_directory): for file in files: if file.startswith("prediction.results"): results_files.append(os.path.join(dirpath, file)) # Consolidate all the results into a list. results = [] for results_file in results_files: # Download each result. with open(results_file, "r") as file: results.extend([json.loads(line) for line in file.readlines()]) # Calculate performance. num_correct = 0 for result in results: label = os.path.basename(result["instance"]).split("_")[0] prediction = result["prediction"]["label"] if label == prediction: num_correct = num_correct + 1 accuracy = num_correct / len(results) logging.info(f"Accuracy: {accuracy*100}%") # Store the boolean result. trigger_pipeline.set_string_custom_property("result", str(accuracy >= threshold))
deep-diver/Continuous-Adaptation-for-Machine-Learning-System-to-Data-Changes
custom_components/batch_prediction_vertex.py
""" This component launches a Batch Prediction job on Vertex AI. Know more about Vertex AI Batch Predictions jobs, go here: https://cloud.google.com/vertex-ai/docs/predictions/batch-predictions. """ from google.cloud import storage from tfx.dsl.component.experimental.annotations import Parameter, InputArtifact from tfx.dsl.component.experimental.decorators import component from tfx.types.standard_artifacts import String import google.cloud.aiplatform as vertex_ai from absl import logging @component def BatchPredictionGen( gcs_source: InputArtifact[String], project: Parameter[str], location: Parameter[str], model_resource_name: Parameter[str], job_display_name: Parameter[str], gcs_destination: Parameter[str], instances_format: Parameter[str] = "file-list", machine_type: Parameter[str] = "n1-standard-2", accelerator_count: Parameter[int] = 0, accelerator_type: Parameter[str] = None, starting_replica_count: Parameter[int] = 1, max_replica_count: Parameter[int] = 1, ): """ gcs_source: A location inside GCS to be used by the Batch Prediction job to get its inputs. Rest of the parameters are explained here: https://git.io/JiUyU. """ storage_client = storage.Client() # Read GCS Source (gcs_source contains the full path of GCS object). # 1-1. get bucketname from gcs_source gcs_source_uri = gcs_source.uri.split("//")[1:][0].split("/") bucketname = gcs_source_uri[0] bucket = storage_client.get_bucket(bucketname) logging.info(f"bucketname: {bucketname}") # 1-2. get object path without the bucket name. objectpath = "/".join(gcs_source_uri[1:]) # 1-3. read the object to get value set by OutputArtifact from FileListGen. blob = bucket.blob(objectpath) logging.info(f"objectpath: {objectpath}") gcs_source = f"gs://{blob.download_as_text()}" # Get Model. vertex_ai.init(project=project, location=location) model = vertex_ai.Model.list( filter=f"display_name={model_resource_name}", order_by="update_time" )[-1] # Launch a Batch Prediction job. logging.info("Starting batch prediction job.") logging.info(f"GCS path where file list is: {gcs_source}") batch_prediction_job = model.batch_predict( job_display_name=job_display_name, instances_format=instances_format, gcs_source=gcs_source, gcs_destination_prefix=gcs_destination, machine_type=machine_type, accelerator_count=accelerator_count, accelerator_type=accelerator_type, starting_replica_count=starting_replica_count, max_replica_count=max_replica_count, sync=True, ) logging.info(batch_prediction_job.display_name) logging.info(batch_prediction_job.resource_name) logging.info(batch_prediction_job.state)
deep-diver/Continuous-Adaptation-for-Machine-Learning-System-to-Data-Changes
custom_components/file_list_gen.py
""" Generate a txt file formatted required by Vertex AI's Batch Prediction There are few options, and this component generate "file list" formatted txt. (https://cloud.google.com/vertex-ai/docs/predictions/batch-predictions) """ import tensorflow as tf from absl import logging from tfx.dsl.component.experimental.decorators import component from tfx.dsl.component.experimental.annotations import Parameter, OutputArtifact from tfx.types.standard_artifacts import String @component def FileListGen( outpath: OutputArtifact[String], gcs_src_bucket: Parameter[str], gcs_src_prefix: Parameter[str] = "", output_filename: Parameter[str] = "test-images.txt", ): """ : param outpath: OutputArtifact to hold where output_filename will be located This will be used in the downstream component, BatchPredictionGen : param gcs_src_bucket: GCS bucket name where the list of raw data is : param gcs_src_prefix: prefix to be added to gcs_src_bucket : param output_filename: output filename whose content is a list of file paths of raw data """ logging.info("FileListGen started") # 1. get the list of data gcs_src_prefix = ( f"{gcs_src_prefix}/" if len(gcs_src_prefix) != 0 else gcs_src_prefix ) img_paths = tf.io.gfile.glob(f"gs://{gcs_src_bucket}/{gcs_src_prefix}*.jpg") logging.info("Successfully retrieve the file(jpg) list from GCS path") # 2. write the list of data in the expected format in Vertex AI Batch Prediction to a local file with open(output_filename, "w", encoding="utf-8") as f: f.writelines("%s\n" % img_path for img_path in img_paths) logging.info( f"Successfully created the file list file({output_filename}) in local storage" ) # 3. upload the local file to GCS location gcs_dst = f"{gcs_src_bucket}/{gcs_src_prefix}{output_filename}" tf.io.gfile.copy(output_filename, f"gs://{gcs_dst}", overwrite=True) logging.info(f"Successfully uploaded the file list ({gcs_dst})") # 4. store the GCS location where the local file is outpath.value = gcs_dst
deep-diver/Continuous-Adaptation-for-Machine-Learning-System-to-Data-Changes
custom_components/span_preparator.py
""" This component is responsible for separating provided samples into training and validation splits. It then converts them to TFRecords and stores those inside a GCS location. Finally, it returns the latest span id calculated from the current samples in `gcs_source_bucket`. """ from tfx.dsl.component.experimental.decorators import component from tfx.dsl.component.experimental.annotations import Parameter from tfx.dsl.component.experimental.annotations import OutputArtifact, InputArtifact from tfx.types.experimental.simple_artifacts import Dataset from absl import logging from datetime import datetime import tensorflow as tf import random import os # Label-mapping. LABEL_DICT = { "airplane": 0, "automobile": 1, "bird": 2, "cat": 3, "deer": 4, "dog": 5, "frog": 6, "horse": 7, "ship": 8, "truck": 9, } # Images are byte-strings. def _bytestring_feature(list_of_bytestrings): return tf.train.Feature(bytes_list=tf.train.BytesList(value=list_of_bytestrings)) # Classes would be integers. def _int_feature(list_of_ints): return tf.train.Feature(int64_list=tf.train.Int64List(value=list_of_ints)) # Function that prepares a record for the tfrecord file # a record contains the image and its label. def to_tfrecord(img_bytes, label): feature = { "image": _bytestring_feature([img_bytes]), "label": _int_feature([label]), } return tf.train.Example(features=tf.train.Features(feature=feature)) def write_tfrecords(filepaths, dest_gcs, tfrecord_filename, new_span, is_train): # For this project, we are serializing the images in one TFRecord only. # For more realistic purposes, this should be sharded. folder = "train" if is_train else "test" with tf.io.TFRecordWriter(tfrecord_filename) as writer: for path in filepaths: image_string = tf.io.read_file(path).numpy() class_name = path.split("/")[-1].split("_")[0] label = LABEL_DICT[class_name] example = to_tfrecord(image_string, label) writer.write(example.SerializeToString()) # Copy over the zipped TFRecord file to the GCS Bucket and # remove the temporary files. logging.info(f"gsutil cp {tfrecord_filename} {dest_gcs}/span-{new_span}/{folder}/") os.system(f"gsutil cp {tfrecord_filename} {dest_gcs}/span-{new_span}/{folder}/") os.remove(tfrecord_filename) @component def SpanPreparator( is_retrain: InputArtifact[Dataset], gcs_source_bucket: Parameter[str], gcs_destination_bucket: Parameter[str], latest_span_id: OutputArtifact[Dataset], gcs_source_prefix: Parameter[str] = "", ): """ :param is_retrain: Boolean to indicate if we are retraining. :param gcs_source_bucket: GCS location where the entry samples are residing. :param gcs_destination_bucket: GCS location where the converted TFRecords will be serialized. :param latest_span_id: Data span. :param gcs_source_prefix: Location prefix. """ if is_retrain.get_string_custom_property("result") == "False": # Get the latest span and determine the new span. last_span_str = tf.io.gfile.glob(f"{gcs_destination_bucket}/span-*")[-1] last_span = int(last_span_str.split("-")[-1]) new_span = last_span + 1 timestamp = datetime.utcnow().strftime("%y%m%d-%H%M%S") # Get images from the provided GCS source. image_paths = tf.io.gfile.glob(f"gs://{gcs_source_bucket}/*.jpg") logging.info(image_paths) random.shuffle(image_paths) # Create train and validation splits. val_split = 0.2 split_index = int(len(image_paths) * (1 - val_split)) training_paths = image_paths[:split_index] validation_paths = image_paths[split_index:] # Write as TFRecords. write_tfrecords( training_paths, gcs_destination_bucket, tfrecord_filename=f"new_training_data_{timestamp}.tfrecord", new_span=new_span, is_train=True, ) write_tfrecords( validation_paths, gcs_destination_bucket, tfrecord_filename=f"new_validation_data_{timestamp}.tfrecord", new_span=new_span, is_train=False, ) logging.info("Removing images from batch prediction bucket.") os.system( f"gsutil mv gs://{gcs_source_bucket}/{gcs_source_prefix} gs://{gcs_source_bucket}/{gcs_source_prefix}_old" ) latest_span_id.set_string_custom_property("latest_span", str(new_span))
deep-diver/Continuous-Adaptation-for-Machine-Learning-System-to-Data-Changes
custom_components/training_pipeline_trigger.py
""" Component responsible for triggering a training job given a pipeline specification. """ import json from google.cloud import storage from kfp.v2.google.client import AIPlatformClient from tfx.dsl.component.experimental.annotations import Parameter, InputArtifact from tfx.dsl.component.experimental.decorators import component from tfx.types.experimental.simple_artifacts import Dataset from absl import logging @component def PipelineTrigger( is_retrain: InputArtifact[Dataset], latest_span_id: InputArtifact[Dataset], pipeline_spec_path: Parameter[str], project_id: Parameter[str], region: Parameter[str], ): """ :param is_retrain: Boolean to indicate if we are retraining. :param latest_span_id: Latest span id to craft training data for the model. :param pipeline_spec_path: Training pipeline specification path. :param project_id: GCP project id. :param region: GCP region. """ if is_retrain.get_string_custom_property("result") == "False": # Check if the pipeline spec exists. storage_client = storage.Client() path_parts = pipeline_spec_path.replace("gs://", "").split("/") bucket_name = path_parts[0] blob_name = "/".join(path_parts[1:]) bucket = storage_client.bucket(bucket_name) blob = storage.Blob(bucket=bucket, name=blob_name) if not blob.exists(storage_client): raise ValueError(f"{pipeline_spec_path} does not exist.") # Initialize Vertex AI API client and submit for pipeline execution. api_client = AIPlatformClient(project_id=project_id, region=region) # Fetch the latest span. latest_span = latest_span_id.get_string_custom_property("latest_span") # Create a training job from pipeline spec. response = api_client.create_run_from_job_spec( pipeline_spec_path, enable_caching=False, parameter_values={ "input-config": json.dumps( { "splits": [ { "name": "train", "pattern": f"span-[{int(latest_span)-1}{latest_span}]/train/*.tfrecord", }, { "name": "val", "pattern": f"span-[{int(latest_span)-1}{latest_span}]/test/*.tfrecord", }, ] } ), "output-config": json.dumps({}), }, ) logging.info(response)
deep-diver/Continuous-Adaptation-for-Machine-Learning-System-to-Data-Changes
notebooks/01_Dataset_Prep.ipynb
from google.colab import auth auth.authenticate_user()TARGET_ROOT_DIR = "cifar10" TARGET_TRAIN_DIR = TARGET_ROOT_DIR + "/span-1/train" TARGET_TEST_DIR = TARGET_ROOT_DIR + "/span-1/test" !mkdir -p {TARGET_TRAIN_DIR} !mkdir -p {TARGET_TEST_DIR}import tensorflow_datasets as tfds # Generate TFRecords with TFDS builder = tfds.builder("cifar10") builder.download_and_prepare()#@title GCS #@markdown You should change these values as per your preferences. The copy operation can take ~5 minutes. BUCKET_PATH = "gs://cifar10-csp-public2" #@param {type:"string"} REGION = "us-central1" #@param {type:"string"} !gsutil mb -l {REGION} {BUCKET_PATH} !gsutil -m cp -r {TARGET_ROOT_DIR}/* {BUCKET_PATH}from tfx import v1 as tfx from tfx.components.example_gen import utilsfrom tfx.proto import example_gen_pb2 _DATA_PATH = "gs://cifar10-csp-public" splits = [ example_gen_pb2.Input.Split(name="train", pattern="span-{SPAN}/train/*"), example_gen_pb2.Input.Split(name="val", pattern="span-{SPAN}/test/*"), ] _, span, version = utils.calculate_splits_fingerprint_span_and_version( _DATA_PATH, splits )span, version
deep-diver/Continuous-Adaptation-for-Machine-Learning-System-to-Data-Changes
notebooks/02_TFX_Training_Pipeline.ipynb
from google.colab import auth auth.authenticate_user()import tensorflow as tf print("TensorFlow version: {}".format(tf.__version__)) from tfx import v1 as tfx print("TFX version: {}".format(tfx.__version__)) import kfp print("KFP version: {}".format(kfp.__version__)) from google.cloud import aiplatform as vertex_ai import osGOOGLE_CLOUD_PROJECT = "gcp-ml-172005" # @param {type:"string"} GOOGLE_CLOUD_REGION = "us-central1" # @param {type:"string"} GCS_BUCKET_NAME = "cifar10-experimental-csp2" # @param {type:"string"} DATA_ROOT = "gs://cifar10-csp-public2" # @param {type:"string"} if not (GOOGLE_CLOUD_PROJECT and GOOGLE_CLOUD_REGION and GCS_BUCKET_NAME): from absl import logging logging.error("Please set all required parameters.")PIPELINE_NAME = "continuous-adaptation-for-data-changes" # Path to various pipeline artifact. PIPELINE_ROOT = "gs://{}/pipeline_root/{}".format(GCS_BUCKET_NAME, PIPELINE_NAME) # Paths for users' Python module. MODULE_ROOT = "gs://{}/pipeline_module/{}".format(GCS_BUCKET_NAME, PIPELINE_NAME) # This is the path where your model will be pushed for serving. SERVING_MODEL_DIR = "gs://{}/serving_model/{}".format(GCS_BUCKET_NAME, PIPELINE_NAME) print("PIPELINE_ROOT: {}".format(PIPELINE_ROOT))_trainer_module_file = 'trainer.py'%%writefile {_trainer_module_file} from typing import List from absl import logging from tensorflow import keras from tfx import v1 as tfx import tensorflow as tf _IMAGE_FEATURES = { "image": tf.io.FixedLenFeature([], tf.string), "label": tf.io.FixedLenFeature([], tf.int64), } _CONCRETE_INPUT = "numpy_inputs" _TRAIN_BATCH_SIZE = 64 _EVAL_BATCH_SIZE = 64 _INPUT_SHAPE = (32, 32, 3) _EPOCHS = 2 def _parse_fn(example): example = tf.io.parse_single_example(example, _IMAGE_FEATURES) image = tf.image.decode_jpeg(example["image"], channels=3) class_label = tf.cast(example["label"], tf.int32) return image, class_label def _input_fn(file_pattern: List[str], batch_size: int) -> tf.data.Dataset: print(f"Reading data from: {file_pattern}") tfrecord_filenames = tf.io.gfile.glob(file_pattern[0] + ".gz") print(tfrecord_filenames) dataset = tf.data.TFRecordDataset(tfrecord_filenames, compression_type="GZIP") dataset = dataset.map(_parse_fn).batch(batch_size) return dataset.repeat() def _make_keras_model() -> tf.keras.Model: """Creates a ResNet50-based model for classifying flowers data. Returns: A Keras Model. """ inputs = keras.Input(shape=_INPUT_SHAPE) base_model = keras.applications.ResNet50( include_top=False, input_shape=_INPUT_SHAPE, pooling="avg" ) base_model.trainable = False x = tf.keras.applications.resnet.preprocess_input(inputs) x = base_model( x, training=False ) # Ensures BatchNorm runs in inference model in this model outputs = keras.layers.Dense(10, activation="softmax")(x) model = keras.Model(inputs, outputs) model.compile( optimizer=keras.optimizers.Adam(), loss=tf.keras.losses.SparseCategoricalCrossentropy(), metrics=[keras.metrics.SparseCategoricalAccuracy()], ) model.summary(print_fn=logging.info) return model def _preprocess(bytes_input): decoded = tf.io.decode_jpeg(bytes_input, channels=3) resized = tf.image.resize(decoded, size=(32, 32)) return resized @tf.function(input_signature=[tf.TensorSpec([None], tf.string)]) def preprocess_fn(bytes_inputs): decoded_images = tf.map_fn( _preprocess, bytes_inputs, dtype=tf.float32, back_prop=False ) return {_CONCRETE_INPUT: decoded_images} def _model_exporter(model: tf.keras.Model): m_call = tf.function(model.call).get_concrete_function( [tf.TensorSpec(shape=[None, 32, 32, 3], dtype=tf.float32, name=_CONCRETE_INPUT)] ) @tf.function(input_signature=[tf.TensorSpec([None], tf.string)]) def serving_fn(bytes_inputs): # This function comes from the Computer Vision book from O'Reilly. labels = tf.constant( [ "airplane", "automobile", "bird", "cat", "deer", "dog", "frog", "horse", "ship", "truck", ], dtype=tf.string, ) images = preprocess_fn(bytes_inputs) probs = m_call(**images) indices = tf.argmax(probs, axis=1) pred_source = tf.gather(params=labels, indices=indices) pred_confidence = tf.reduce_max(probs, axis=1) return {"label": pred_source, "confidence": pred_confidence} return serving_fn def run_fn(fn_args: tfx.components.FnArgs): print(fn_args) train_dataset = _input_fn(fn_args.train_files, batch_size=_TRAIN_BATCH_SIZE) eval_dataset = _input_fn(fn_args.eval_files, batch_size=_EVAL_BATCH_SIZE) model = _make_keras_model() model.fit( train_dataset, steps_per_epoch=fn_args.train_steps, validation_data=eval_dataset, validation_steps=fn_args.eval_steps, epochs=_EPOCHS, ) _, acc = model.evaluate(eval_dataset, steps=fn_args.eval_steps) logging.info(f"Validation accuracy: {round(acc * 100, 2)}%") # The result of the training should be saved in `fn_args.serving_model_dir` # directory. tf.saved_model.save( model, fn_args.serving_model_dir, signatures={"serving_default": _model_exporter(model)}, )os.path.join(MODULE_ROOT, _trainer_module_file)_vertex_uploader_module_file = "vertex_uploader.py" _vertex_deployer_module_file = "vertex_deployer.py"%%writefile {_vertex_uploader_module_file} import os import tensorflow as tf from tfx.dsl.component.experimental.decorators import component from tfx.dsl.component.experimental.annotations import Parameter from tfx.types.standard_artifacts import String from google.cloud import aiplatform as vertex_ai from tfx import v1 as tfx from absl import logging @component def VertexUploader( project: Parameter[str], region: Parameter[str], model_display_name: Parameter[str], pushed_model_location: Parameter[str], serving_image_uri: Parameter[str], uploaded_model: tfx.dsl.components.OutputArtifact[String], ): vertex_ai.init(project=project, location=region) pushed_model_dir = os.path.join( pushed_model_location, tf.io.gfile.listdir(pushed_model_location)[-1] ) logging.info(f"Model registry location: {pushed_model_dir}") vertex_model = vertex_ai.Model.upload( display_name=model_display_name, artifact_uri=pushed_model_dir, serving_container_image_uri=serving_image_uri, parameters_schema_uri=None, instance_schema_uri=None, explanation_metadata=None, explanation_parameters=None, ) uploaded_model.set_string_custom_property( "model_resource_name", str(vertex_model.resource_name) ) logging.info(f"Model resource: {str(vertex_model.resource_name)}")%%writefile {_vertex_deployer_module_file} from tfx.dsl.component.experimental.decorators import component from tfx.dsl.component.experimental.annotations import Parameter from tfx.types.standard_artifacts import String from google.cloud import aiplatform as vertex_ai from tfx import v1 as tfx from absl import logging @component def VertexDeployer( project: Parameter[str], region: Parameter[str], model_display_name: Parameter[str], deployed_model_display_name: Parameter[str], ): logging.info(f"Endpoint display: {deployed_model_display_name}") vertex_ai.init(project=project, location=region) endpoints = vertex_ai.Endpoint.list( filter=f"display_name={deployed_model_display_name}", order_by="update_time" ) if len(endpoints) > 0: logging.info(f"Endpoint {deployed_model_display_name} already exists.") endpoint = endpoints[-1] else: endpoint = vertex_ai.Endpoint.create(deployed_model_display_name) model = vertex_ai.Model.list( filter=f"display_name={model_display_name}", order_by="update_time" )[-1] endpoint = vertex_ai.Endpoint.list( filter=f"display_name={deployed_model_display_name}", order_by="update_time" )[-1] deployed_model = endpoint.deploy( model=model, # Syntax from here: https://git.io/JBQDP traffic_split={"0": 100}, machine_type="n1-standard-4", min_replica_count=1, max_replica_count=1, ) logging.info(f"Model deployed to: {deployed_model}")DATASET_DISPLAY_NAME = "cifar10" VERSION = "tfx-1-2-0" TFX_IMAGE_URI = f"gcr.io/{GOOGLE_CLOUD_PROJECT}/{DATASET_DISPLAY_NAME}:{VERSION}" print(f"URI of the custom image: {TFX_IMAGE_URI}")%%writefile Dockerfile FROM gcr.io/tfx-oss-public/tfx:1.2.0 RUN mkdir -p custom_components COPY custom_components/* ./custom_components/ RUN pip install --upgrade google-cloud-aiplatform# Specify training worker configurations. To minimize costs we can even specify two # different configurations: a beefier machine for the Endpoint model and slightly less # powerful machine for the mobile model. TRAINING_JOB_SPEC = { "project": GOOGLE_CLOUD_PROJECT, "worker_pool_specs": [ { "machine_spec": { "machine_type": "n1-standard-4", "accelerator_type": "NVIDIA_TESLA_K80", "accelerator_count": 1, }, "replica_count": 1, "container_spec": { "image_uri": "gcr.io/tfx-oss-public/tfx:{}".format(tfx.__version__), }, } ], }SERVING_JOB_SPEC = { "endpoint_name": PIPELINE_NAME.replace("-", "_"), # '-' is not allowed. "project_id": GOOGLE_CLOUD_PROJECT, "min_replica_count": 1, "max_replica_count": 1, "machine_type": "n1-standard-2", }from datetime import datetime TIMESTAMP = datetime.now().strftime("%Y%m%d%H%M%S")import tfxfrom tfx.orchestration import data_types from tfx import v1 as tfx from tfx.proto import example_gen_pb2, range_config_pb2 from tfx.components.example_gen import utils from custom_components.vertex_uploader import VertexUploader from custom_components.vertex_deployer import VertexDeployer def _create_pipeline( input_config: data_types.RuntimeParameter, output_config: data_types.RuntimeParameter, pipeline_name: str, pipeline_root: str, data_root: str, serving_model_dir: str, trainer_module: str, project_id: str, region: str, ) -> tfx.dsl.Pipeline: """Creates a three component flowers pipeline with TFX.""" example_gen = tfx.components.ImportExampleGen( input_base=data_root, input_config=input_config, output_config=output_config ) # Trainer trainer = tfx.extensions.google_cloud_ai_platform.Trainer( module_file=trainer_module, examples=example_gen.outputs["examples"], train_args=tfx.proto.TrainArgs(splits=["train"], num_steps=50000 // 64), eval_args=tfx.proto.EvalArgs(splits=["val"], num_steps=10000 // 64), custom_config={ tfx.extensions.google_cloud_ai_platform.ENABLE_VERTEX_KEY: True, tfx.extensions.google_cloud_ai_platform.VERTEX_REGION_KEY: region, tfx.extensions.google_cloud_ai_platform.TRAINING_ARGS_KEY: TRAINING_JOB_SPEC, "use_gpu": True, }, ).with_id("trainer") # Pushes the model to a filesystem destination. pushed_model_location = os.path.join(serving_model_dir, "resnet50") resnet_pusher = tfx.components.Pusher( model=trainer.outputs["model"], push_destination=tfx.proto.PushDestination( filesystem=tfx.proto.PushDestination.Filesystem( base_directory=pushed_model_location ) ), ).with_id("resnet_pusher") # Vertex AI upload. model_display_name = "resnet_cifar_latest" uploader = VertexUploader( project=project_id, region=region, model_display_name=model_display_name, pushed_model_location=pushed_model_location, serving_image_uri="us-docker.pkg.dev/vertex-ai/prediction/tf2-cpu.2-5:latest", ).with_id("vertex_uploader") uploader.add_upstream_node(resnet_pusher) # Create an endpoint. deployer = VertexDeployer( project=project_id, region=region, model_display_name=model_display_name, deployed_model_display_name=model_display_name + "_" + TIMESTAMP, ).with_id("vertex_deployer") deployer.add_upstream_node(uploader) components = [ example_gen, trainer, resnet_pusher, uploader, deployer, ] return tfx.dsl.Pipeline( pipeline_name=pipeline_name, pipeline_root=pipeline_root, components=components, enable_cache=True, )import os PIPELINE_DEFINITION_FILE = PIPELINE_NAME + "_pipeline.json" # Important: We need to pass the custom Docker image URI to the # `KubeflowV2DagRunnerConfig` to take effect. runner = tfx.orchestration.experimental.KubeflowV2DagRunner( config=tfx.orchestration.experimental.KubeflowV2DagRunnerConfig( default_image=TFX_IMAGE_URI ), output_filename=PIPELINE_DEFINITION_FILE, ) _ = runner.run( _create_pipeline( input_config=tfx.dsl.experimental.RuntimeParameter( name="input-config", default='{"input_config": {"splits": [{"name":"train", "pattern":"span-1/train/tfrecord"}, {"name":"val", "pattern":"span-1/test/tfrecord"}]}}', ptype=str, ), output_config=tfx.dsl.experimental.RuntimeParameter( name="output-config", default="{}", ptype=str, ), pipeline_name=PIPELINE_NAME, pipeline_root=PIPELINE_ROOT, data_root=DATA_ROOT, serving_model_dir=SERVING_MODEL_DIR, trainer_module=os.path.join(MODULE_ROOT, _trainer_module_file), project_id=GOOGLE_CLOUD_PROJECT, region=GOOGLE_CLOUD_REGION, ) )from kfp.v2.google import client pipelines_client = client.AIPlatformClient( project_id=GOOGLE_CLOUD_PROJECT, region=GOOGLE_CLOUD_REGION, )import json from tfx.orchestration import data_types _ = pipelines_client.create_run_from_job_spec( PIPELINE_DEFINITION_FILE, enable_caching=False, parameter_values={ "input-config": json.dumps( { "splits": [ {"name": "train", "pattern": "span-[12]/train/*.tfrecord"}, {"name": "val", "pattern": "span-[12]/test/*.tfrecord"}, ] } ), "output-config": json.dumps({}), }, )
deep-diver/Continuous-Adaptation-for-Machine-Learning-System-to-Data-Changes
notebooks/03_Batch_Prediction_Pipeline.ipynb
from google.colab import auth auth.authenticate_user()# @title from fastdot.core import * tfx_components = [ "FileListGen", "BatchPredictionGen", "PerformanceEvaluator", "SpanPreparator", "PipelineTrigger", ] block = "TFX Component Workflow" g = graph_items(seq_cluster(tfx_components, block)) g_file_list_gen_module_file = "file_list_gen.py"%%writefile {_file_list_gen_module_file} import tfx from tfx.dsl.component.experimental.decorators import component from tfx.dsl.component.experimental.annotations import Parameter from tfx.dsl.component.experimental.annotations import OutputArtifact from tfx.types.standard_artifacts import String from google.cloud import storage from absl import logging @component def FileListGen( outpath: OutputArtifact[String], project: Parameter[str], gcs_source_bucket: Parameter[str], gcs_source_prefix: Parameter[str] = "", output_filename: Parameter[str] = "test-images.txt", ): logging.info("FileListGen started") client = storage.Client(project=project) bucket = client.get_bucket(gcs_source_bucket) blobs = bucket.list_blobs(prefix=gcs_source_prefix) logging.info("Successfully retrieve the file(jpg) list from GCS path") f = open(output_filename, "w") for blob in blobs: if blob.name.split(".")[-1] == "jpg": prefix = "" if gcs_source_prefix != "": prefix = f"/{gcs_source_prefix}" line = f"gs://{gcs_source_bucket}{prefix}/{blob.name}\n" f.write(line) f.close() logging.info( f"Successfully created the file list file({output_filename}) in local storage" ) prefix = "" if gcs_source_prefix != "": prefix = f"{gcs_source_prefix}/" blob = bucket.blob(f"{prefix}{output_filename}") blob.upload_from_filename(output_filename) logging.info(f"Successfully uploaded the file list ({prefix}{output_filename})") outpath.value = gcs_source_bucket + "/" + prefix + output_filename_batch_pred_module_file = 'batch_prediction_vertex.py'%%writefile {_batch_pred_module_file} from google.cloud import storage from tfx.dsl.component.experimental.annotations import Parameter, InputArtifact from tfx.dsl.component.experimental.decorators import component from tfx.types import artifact_utils from tfx.types.standard_artifacts import String import google.cloud.aiplatform as vertex_ai from typing import Union, Sequence from absl import logging @component def BatchPredictionGen( gcs_source: InputArtifact[String], project: Parameter[str], location: Parameter[str], model_resource_name: Parameter[str], job_display_name: Parameter[str], gcs_destination: Parameter[str], instances_format: Parameter[str] = "file-list", machine_type: Parameter[str] = "n1-standard-2", accelerator_count: Parameter[int] = 0, accelerator_type: Parameter[str] = None, starting_replica_count: Parameter[int] = 1, max_replica_count: Parameter[int] = 1, ): storage_client = storage.Client() # Read GCS Source (gcs_source contains the full path of GCS object) # 1-1. get bucketname from gcs_source gcs_source_uri = gcs_source.uri.split("//")[1:][0].split("/") bucketname = gcs_source_uri[0] bucket = storage_client.get_bucket(bucketname) logging.info(f"bucketname: {bucketname}") # 1-2. get object path without the bucketname objectpath = "/".join(gcs_source_uri[1:]) # 1-3. read the object to get value set by OutputArtifact from FileListGen blob = bucket.blob(objectpath) logging.info(f"objectpath: {objectpath}") gcs_source = f"gs://{blob.download_as_text()}" # Get Model vertex_ai.init(project=project, location=location) model = vertex_ai.Model.list( filter=f"display_name={model_resource_name}", order_by="update_time" )[-1] # Batch Predictions logging.info("Starting batch prediction job.") logging.info(f"GCS path where file list is: {gcs_source}") batch_prediction_job = model.batch_predict( job_display_name=job_display_name, instances_format=instances_format, gcs_source=gcs_source, gcs_destination_prefix=gcs_destination, machine_type=machine_type, accelerator_count=accelerator_count, accelerator_type=accelerator_type, starting_replica_count=starting_replica_count, max_replica_count=max_replica_count, sync=True, ) logging.info(batch_prediction_job.display_name) logging.info(batch_prediction_job.resource_name) logging.info(batch_prediction_job.state)_evaluator_module_file = 'batch_pred_evaluator.py'%%writefile {_evaluator_module_file} # Reference: https://bit.ly/vertex-batch from tfx.dsl.component.experimental.annotations import Parameter from tfx.dsl.component.experimental.annotations import OutputArtifact from tfx.dsl.component.experimental.decorators import component from tfx.types.experimental.simple_artifacts import Dataset from absl import logging import os import json @component def PerformanceEvaluator( gcs_destination: Parameter[str], local_directory: Parameter[str], threshold: Parameter[float], trigger_pipeline: OutputArtifact[Dataset], ): full_gcs_results_dir = f"{gcs_destination}/{local_directory}" # Create missing directories. os.makedirs(local_directory, exist_ok=True) # Get the Cloud Storage paths for each result. os.system(f"gsutil -m cp -r {full_gcs_results_dir} {local_directory}") # Get most recently modified directory. latest_directory = max( [os.path.join(local_directory, d) for d in os.listdir(local_directory)], key=os.path.getmtime, ) # Get downloaded results in directory. results_files = [] for dirpath, subdirs, files in os.walk(latest_directory): for file in files: if file.startswith("prediction.results"): results_files.append(os.path.join(dirpath, file)) # Consolidate all the results into a list. results = [] for results_file in results_files: # Download each result. with open(results_file, "r") as file: results.extend([json.loads(line) for line in file.readlines()]) # Calculate performance. num_correct = 0 for result in results: label = os.path.basename(result["instance"]).split("_")[0] prediction = result["prediction"]["label"] if label == prediction: num_correct = num_correct + 1 accuracy = num_correct / len(results) logging.info(f"Accuracy: {accuracy*100}%") trigger_pipeline.set_string_custom_property("result", str(accuracy >= threshold))_span_preparator_module_file = 'span_preparator.py'%%writefile {_span_preparator_module_file} import tfx from tfx.dsl.component.experimental.decorators import component from tfx.dsl.component.experimental.annotations import Parameter from tfx.dsl.component.experimental.annotations import OutputArtifact, InputArtifact from tfx.types.experimental.simple_artifacts import Dataset from google.cloud import storage from absl import logging from datetime import datetime import tensorflow as tf import random import gzip import os # Label-mapping. LABEL_DICT = { "airplane": 0, "automobile": 1, "bird": 2, "cat": 3, "deer": 4, "dog": 5, "frog": 6, "horse": 7, "ship": 8, "truck": 9, } # Images are byte-strings. def _bytestring_feature(list_of_bytestrings): return tf.train.Feature(bytes_list=tf.train.BytesList(value=list_of_bytestrings)) # Classes would be integers. def _int_feature(list_of_ints): return tf.train.Feature(int64_list=tf.train.Int64List(value=list_of_ints)) # Function that prepares a record for the tfrecord file # a record contains the image and its label. def to_tfrecord(img_bytes, label): feature = { "image": _bytestring_feature([img_bytes]), "label": _int_feature([label]), } return tf.train.Example(features=tf.train.Features(feature=feature)) def write_tfrecords(filepaths, dest_gcs, tfrecord_filename, new_span, is_train): # For this project, we are serializing the images in one TFRecord only. # For more realistic purposes, this should be sharded. folder = "train" if is_train else "test" with tf.io.TFRecordWriter(tfrecord_filename) as writer: for path in filepaths: image_string = tf.io.read_file(path).numpy() class_name = path.split("/")[-1].split("_")[0] label = LABEL_DICT[class_name] example = to_tfrecord(image_string, label) writer.write(example.SerializeToString()) # Copy over the zipped TFRecord file to the GCS Bucket and # remove the temporary files. logging.info(f"gsutil cp {tfrecord_filename} {dest_gcs}/span-{new_span}/{folder}/") os.system(f"gsutil cp {tfrecord_filename} {dest_gcs}/span-{new_span}/{folder}/") os.remove(tfrecord_filename) @component def SpanPreparator( is_retrain: InputArtifact[Dataset], gcs_source_bucket: Parameter[str], gcs_destination_bucket: Parameter[str], latest_span_id: OutputArtifact[Dataset], gcs_source_prefix: Parameter[str] = "", ): if is_retrain.get_string_custom_property("result") == "False": last_span_str = tf.io.gfile.glob(f"{gcs_destination_bucket}/span-*")[-1] last_span = int(last_span_str.split("-")[-1]) new_span = last_span + 1 timestamp = datetime.utcnow().strftime("%y%m%d-%H%M%S") image_paths = tf.io.gfile.glob(f"gs://{gcs_source_bucket}/*.jpg") logging.info(image_paths) random.shuffle(image_paths) val_split = 0.2 split_index = int(len(image_paths) * (1 - val_split)) training_paths = image_paths[:split_index] validation_paths = image_paths[split_index:] write_tfrecords( training_paths, gcs_destination_bucket, tfrecord_filename=f"new_training_data_{timestamp}.tfrecord", new_span=new_span, is_train=True, ) write_tfrecords( validation_paths, gcs_destination_bucket, tfrecord_filename=f"new_validation_data_{timestamp}.tfrecord", new_span=new_span, is_train=False, ) logging.info("Removing images from batch prediction bucket.") os.system( f"gsutil mv gs://{gcs_source_bucket}/{gcs_source_prefix} gs://{gcs_source_bucket}/{gcs_source_prefix}_old" ) # os.system(f"gsutil rm -rf gs://{gcs_source_bucket}/*") latest_span_id.set_string_custom_property("latest_span", str(new_span))_pipeline_trigger_module_file = 'training_pipeline_trigger.py'%%writefile {_pipeline_trigger_module_file} import json from google.cloud import storage from kfp.v2.google.client import AIPlatformClient from tfx.dsl.component.experimental.annotations import Parameter, InputArtifact from tfx.dsl.component.experimental.decorators import component from tfx.types.experimental.simple_artifacts import Dataset from absl import logging @component def PipelineTrigger( is_retrain: InputArtifact[Dataset], latest_span_id: InputArtifact[Dataset], pipeline_spec_path: Parameter[str], project_id: Parameter[str], region: Parameter[str], ): if is_retrain.get_string_custom_property('result') == 'False': # Check if the pipeline spec exists. storage_client = storage.Client() path_parts = pipeline_spec_path.replace("gs://", "").split("/") bucket_name = path_parts[0] blob_name = "/".join(path_parts[1:]) bucket = storage_client.bucket(bucket_name) blob = storage.Blob(bucket=bucket, name=blob_name) if not blob.exists(storage_client): raise ValueError(f"{pipeline_spec_path} does not exist.") # Initialize Vertex AI API client and submit for pipeline execution. api_client = AIPlatformClient(project_id=project_id, region=region) # Fetch the latest span. latest_span = latest_span_id.get_string_custom_property('latest_span') # Create a training job from pipeline spec. response = api_client.create_run_from_job_spec(pipeline_spec_path, enable_caching=False, parameter_values={ 'input-config': json.dumps({ 'splits': [ {'name': 'train', 'pattern': f'span-[{int(latest_span)-1}{latest_span}]/train/*.tfrecord'}, {'name': 'val', 'pattern': f'span-[{int(latest_span)-1}{latest_span}]/test/*.tfrecord'} ] }), 'output-config': json.dumps({}) }) logging.info(response)# This bucket will be responsible for storing the pipeline related artifacts. GOOGLE_CLOUD_PROJECT = "gcp-ml-172005" # @param {type:"string"} GOOGLE_CLOUD_REGION = "us-central1" GCS_BUCKET_NAME = "cifar10-experimental-csp2" # @param {type:"string"} MODEL_RESOURCE_NAME = "resnet_cifar_latest" # @param {type: "string"} TEST_FILENAME = "test-images.txt" # @param {type:"string"} TEST_GCS_BUCKET = "batch-prediction-collection-3" # @param {type:"string"} TEST_GCS_PREFIX = "" # @param {type: "string"} TRAINING_PIPELINE_SPEC = "gs://cifar10-experimental-csp2/pipeline_root/continuous-adaptation-for-data-changes/continuous-adaptation-for-data-changes_pipeline.json" # @param {type: "string"} TRAINING_DATA_PATH = "gs://cifar10-csp-public2" # @param {type: "string"} if not (GOOGLE_CLOUD_PROJECT and GOOGLE_CLOUD_REGION and GCS_BUCKET_NAME): from absl import logging logging.error("Please set all required parameters.")PIPELINE_NAME = 'continuous-adaptation-for-data-changes-batch' # Path to various pipeline artifact. PIPELINE_ROOT = 'gs://{}/pipeline_root/{}'.format( GCS_BUCKET_NAME, PIPELINE_NAME) print('PIPELINE_ROOT: {}'.format(PIPELINE_ROOT))DISPLAY_NAME = "batch-predictions-pipeline" VERSION = "tfx-1-2-0-34" TFX_IMAGE_URI = f"gcr.io/{GOOGLE_CLOUD_PROJECT}/{DISPLAY_NAME}:{VERSION}" print(f"URI of the custom image: {TFX_IMAGE_URI}")%%writefile Dockerfile FROM gcr.io/tfx-oss-public/tfx:1.2.0 RUN mkdir -p custom_components COPY custom_components/* ./custom_components/ RUN pip install --upgrade google-cloud-aiplatform google-cloud-storage kfp==1.6.1from datetime import datetime TIMESTAMP = datetime.now().strftime("%Y%m%d%H%M%S")from tfx.orchestration import data_types from tfx import v1 as tfx from tfx.orchestration.pipeline import Pipeline from custom_components.file_list_gen import FileListGen from custom_components.batch_prediction_vertex import BatchPredictionGen from custom_components.batch_pred_evaluator import PerformanceEvaluator from custom_components.span_preparator import SpanPreparator from custom_components.training_pipeline_trigger import PipelineTrigger def _create_pipeline( pipeline_name: str, pipeline_root: str, data_gcs_bucket: str, data_gcs_prefix: data_types.RuntimeParameter, batch_job_gcs: str, job_display_name: str, model_resource_name: str, project_id: str, region: str, threshold: float, data_gcs_destination: str, training_pipeline_spec: str, ) -> Pipeline: # Generate a file list for batch preditions. # More details on the structure of this file here: # https://bit.ly/3BzfHVu. filelist_gen = FileListGen( project=project_id, gcs_source_bucket=data_gcs_bucket, gcs_source_prefix=data_gcs_prefix, ).with_id("filelist_gen") # Submit a batch prediction job. batch_pred_component = BatchPredictionGen( project=project_id, location=region, job_display_name=job_display_name, model_resource_name=model_resource_name, gcs_source=filelist_gen.outputs["outpath"], gcs_destination=f"gs://{batch_job_gcs}/results/", accelerator_count=0, accelerator_type=None, ).with_id("bulk_inferer_vertex") batch_pred_component.add_upstream_node(filelist_gen) # Evaluate the performance of the predictions. # In a real-world project, this evaluation takes place # separately, typically with the help of domain experts. final_gcs_destination = f"gs://{batch_job_gcs}/results/" evaluator = PerformanceEvaluator( gcs_destination=f'gs://{final_gcs_destination.split("/")[2]}', local_directory=final_gcs_destination.split("/")[-2], threshold=threshold, ).with_id("batch_prediction_evaluator") evaluator.add_upstream_node(batch_pred_component) span_preparator = SpanPreparator( is_retrain=evaluator.outputs["trigger_pipeline"], gcs_source_bucket=data_gcs_bucket, gcs_source_prefix=data_gcs_prefix, gcs_destination_bucket=data_gcs_destination, ).with_id("span_preparator") span_preparator.add_upstream_node(evaluator) trigger = PipelineTrigger( is_retrain=evaluator.outputs["trigger_pipeline"], latest_span_id=span_preparator.outputs["latest_span_id"], pipeline_spec_path=training_pipeline_spec, project_id=project_id, region=region, ).with_id("training_pipeline_trigger") trigger.add_upstream_node(span_preparator) components = [ filelist_gen, batch_pred_component, evaluator, span_preparator, trigger, ] return Pipeline( pipeline_name=pipeline_name, pipeline_root=pipeline_root, components=components, enable_cache=True, )import os import tfx from tfx.orchestration import data_types from tfx.orchestration.kubeflow.v2.kubeflow_v2_dag_runner import KubeflowV2DagRunner from tfx.orchestration.kubeflow.v2.kubeflow_v2_dag_runner import ( KubeflowV2DagRunnerConfig, ) PIPELINE_DEFINITION_FILE = PIPELINE_NAME + "_pipeline.json" THRESHOLD = 0.9 # Important: We need to pass the custom Docker image URI to the # `KubeflowV2DagRunnerConfig` to take effect. runner = KubeflowV2DagRunner( config=KubeflowV2DagRunnerConfig(default_image=TFX_IMAGE_URI), output_filename=PIPELINE_DEFINITION_FILE, ) _ = runner.run( _create_pipeline( pipeline_name=PIPELINE_NAME, pipeline_root=PIPELINE_ROOT, data_gcs_bucket=TEST_GCS_BUCKET, data_gcs_prefix=data_types.RuntimeParameter( name="data_gcs_prefix", default="", ptype=str ), batch_job_gcs=GCS_BUCKET_NAME, job_display_name=f"{MODEL_RESOURCE_NAME}_{TIMESTAMP}", project_id=GOOGLE_CLOUD_PROJECT, region=GOOGLE_CLOUD_REGION, model_resource_name=MODEL_RESOURCE_NAME, threshold=THRESHOLD, data_gcs_destination=TRAINING_DATA_PATH, training_pipeline_spec=TRAINING_PIPELINE_SPEC, ) )from kfp.v2.google import client pipelines_client = client.AIPlatformClient( project_id=GOOGLE_CLOUD_PROJECT, region=GOOGLE_CLOUD_REGION, ) _ = pipelines_client.create_run_from_job_spec( PIPELINE_DEFINITION_FILE, enable_caching=False, parameter_values={"data_gcs_prefix": "2021-10"}, )
deep-diver/Continuous-Adaptation-for-Machine-Learning-System-to-Data-Changes
notebooks/04_Cloud_Scheduler_Trigger.ipynb
from google.colab import auth auth.authenticate_user()GOOGLE_CLOUD_PROJECT = "gcp-ml-172005" # @param {type:"string"} GOOGLE_CLOUD_REGION = "us-central1" GCS_BUCKET_NAME = "cifar10-experimental-csp2" # @param {type:"string"} PIPELINE_NAME = "continuous-adaptation-for-data-changes-batch" # @param {type:"string"} PIPELINE_ROOT = "gs://{}/pipeline_root/{}".format(GCS_BUCKET_NAME, PIPELINE_NAME) PIPELINE_LOCATION = f"{PIPELINE_ROOT}/{PIPELINE_NAME}_pipeline.json" PUBSUB_TOPIC = f"trigger-{PIPELINE_NAME}" SCHEDULER_JOB_NAME = f"scheduler-job-{PUBSUB_TOPIC}" IMAGE_LOCATION_BUCKET = "batch-prediction-collection-3" # @param {type:"string"}IMAGE_LOCATION_BUCKET_cloud_function_dep = "cloud_function/requirements.txt"%%writefile {_cloud_function_dep} kfp==1.6.2 google-cloud-aiplatform google-cloud-storage_cloud_function_file = "cloud_function/main.py"%%writefile {_cloud_function_file} import os import re import json import logging import base64 from datetime import datetime from kfp.v2.google.client import AIPlatformClient from google.cloud import storage def get_number_of_images(storage_client, bucket, latest_directory): blobs = storage_client.list_blobs(bucket, prefix=latest_directory) count = 0 for blob in blobs: if blob.name.split(".")[-1] == "jpg": count = count + 1 return count def is_there_enough_images(storage_client, bucket, latest_directory, threshold): number_of_images = get_number_of_images(storage_client, bucket, latest_directory) print(f"number of images = {number_of_images}") return number_of_images >= threshold def get_latest_directory(storage_client, bucket): blobs = storage_client.list_blobs(bucket) folders = list( set( [ os.path.dirname(blob.name) for blob in blobs if bool( re.match( "[1-9][0-9][0-9][0-9]-[0-1][0-9]", os.path.dirname(blob.name) ) ) is True ] ) ) folders.sort(key=lambda date: datetime.strptime(date, "%Y-%m")) print(folders[0]) return folders[0] def trigger_pipeline(event, context): # Parse the environment variables. project = os.getenv("PROJECT") region = os.getenv("REGION") gcs_pipeline_file_location = os.getenv("GCS_PIPELINE_FILE_LOCATION") gcs_image_bucket = os.getenv("GCS_IMAGE_BUCKET") print(project) print(region) print(gcs_pipeline_file_location) print(gcs_image_bucket) threshold = 100 # Check if the pipeline file exists in the provided GCS Bucket. storage_client = storage.Client() latest_directory = get_latest_directory(storage_client, gcs_image_bucket) if is_there_enough_images( storage_client, gcs_image_bucket, latest_directory, threshold ): path_parts = gcs_pipeline_file_location.replace("gs://", "").split("/") pipeline_bucket = path_parts[0] pipeline_blob = "/".join(path_parts[1:]) pipeline_bucket = storage_client.bucket(pipeline_bucket) blob = storage.Blob(bucket=pipeline_bucket, name=pipeline_blob) if not blob.exists(storage_client): raise ValueError(f"{gcs_pipeline_file_location} does not exist.") # Initialize Vertex AI API client and submit for pipeline execution. api_client = AIPlatformClient(project_id=project, region=region) response = api_client.create_run_from_job_spec( job_spec_path=gcs_pipeline_file_location, parameter_values={"data_gcs_prefix": latest_directory}, enable_caching=True, ) logging.info(response)ENV_VARS=f"""\ PROJECT={GOOGLE_CLOUD_PROJECT},\ REGION={GOOGLE_CLOUD_REGION},\ GCS_PIPELINE_FILE_LOCATION={PIPELINE_LOCATION},\ GCS_IMAGE_BUCKET={IMAGE_LOCATION_BUCKET} """ !echo {ENV_VARS}BUCKET = f'gs://{GCS_BUCKET_NAME}' CLOUD_FUNCTION_NAME = f'trigger-{PIPELINE_NAME}-fn' !gcloud functions deploy {CLOUD_FUNCTION_NAME} \ --region={GOOGLE_CLOUD_REGION} \ --trigger-topic={PUBSUB_TOPIC} \ --runtime=python37 \ --source=cloud_function\ --entry-point=trigger_pipeline\ --stage-bucket={BUCKET}\ --update-env-vars={ENV_VARS}import IPython cloud_fn_url = f"https://console.cloud.google.com/functions/details/{GOOGLE_CLOUD_REGION}/{CLOUD_FUNCTION_NAME}" html = ( f'See the Cloud Function details <a href="{cloud_fn_url}" target="_blank">here</a>.' ) IPython.display.display(IPython.display.HTML(html))
deep-diver/Continuous-Adaptation-for-Machine-Learning-System-to-Data-Changes
notebooks/98_Batch_Prediction_Test.ipynb
from google.colab import auth auth.authenticate_user()GOOGLE_CLOUD_PROJECT = "central-hangar-321813" # @param {type:"string"} GOOGLE_CLOUD_REGION = "us-central1" # @param {type:"string"} MODEL_NAME = "resnet_cifar_latest" # @param {type:"string"} TEST_FILENAME = "test-images.txt" # @param {type:"string"} TEST_GCS_BUCKET = "gs://batch-prediction-collection" # @param {type:"string"} TEST_LOCAL_PATH = "Continuous-Adaptation-for-Machine-Learning-System-to-Data-Changes/notebooks/test-images" # @param {type:"string"}from os import listdir test_files = listdir(TEST_LOCAL_PATH) test_filesf = open(TEST_FILENAME, "w") for filename in test_files: f.write(f"{TEST_GCS_BUCKET}/{filename}\n") f.close()import google.cloud.aiplatform as aiplatform from typing import Union, Sequence def create_batch_prediction_job_dedicated_resources_sample( project: str, location: str, model_resource_name: str, job_display_name: str, gcs_source: Union[str, Sequence[str]], gcs_destination: str, instances_format: str = "file-list", machine_type: str = "n1-standard-2", accelerator_count: int = 1, accelerator_type: str = "NVIDIA_TESLA_K80", starting_replica_count: int = 1, max_replica_count: int = 1, sync: bool = True, ): aiplatform.init(project=project, location=location) my_model = aiplatform.Model(model_resource_name) batch_prediction_job = my_model.batch_predict( job_display_name=job_display_name, instances_format=instances_format, gcs_source=gcs_source, gcs_destination_prefix=gcs_destination, machine_type=machine_type, accelerator_count=accelerator_count, accelerator_type=accelerator_type, starting_replica_count=starting_replica_count, max_replica_count=max_replica_count, sync=sync, ) batch_prediction_job.wait() print(batch_prediction_job.display_name) print(batch_prediction_job.resource_name) print(batch_prediction_job.state) return batch_prediction_jobfrom datetime import datetime TIMESTAMP = datetime.now().strftime("%Y%m%d%H%M%S")create_batch_prediction_job_dedicated_resources_sample( project=GOOGLE_CLOUD_PROJECT, location=GOOGLE_CLOUD_REGION, model_resource_name="2008244793993330688", job_display_name=f"{MODEL_NAME}-{TIMESTAMP}", gcs_source=[f"{TEST_GCS_BUCKET}/{TEST_FILENAME}"], gcs_destination=f"{TEST_GCS_BUCKET}/results/", accelerator_type=None, accelerator_count=None, )import os import json RESULTS_DIRECTORY = "results" RESULTS_DIRECTORY_FULL = f'{TEST_GCS_BUCKET}/{RESULTS_DIRECTORY}' # Create missing directories os.makedirs(RESULTS_DIRECTORY, exist_ok=True) # Get the Cloud Storage paths for each result !gsutil -m cp -r $RESULTS_DIRECTORY_FULL $RESULTS_DIRECTORY # Get most recently modified directory latest_directory = max( [ os.path.join(RESULTS_DIRECTORY, d) for d in os.listdir(RESULTS_DIRECTORY) ], key=os.path.getmtime, ) # Get downloaded results in directory results_files = [] for dirpath, subdirs, files in os.walk(latest_directory): for file in files: if file.startswith("prediction.results"): results_files.append(os.path.join(dirpath, file)) # Consolidate all the results into a list results = [] for results_file in results_files: # Download each result with open(results_file, "r") as file: results.extend([json.loads(line) for line in file.readlines()])resultsnum_correct = 0 for result in results: label = os.path.basename(result["instance"]).split("_")[0] prediction = result["prediction"]["label"] print(f"label({label})/prediction({prediction})") if label == prediction: num_correct = num_correct + 1 print() print(f"number of results: {len(results)}") print(f"number of correct: {num_correct}") print(f"Accuracy: {num_correct/len(results)}")
deep-diver/mlops-hf-tf-vision-models
advanced_part1/kubeflow_runner.py
from absl import logging from tfx import v1 as tfx from tfx.orchestration.kubeflow.v2 import kubeflow_v2_dag_runner as runner from pipeline import configs from pipeline import kubeflow_pipeline def run(): runner_config = runner.KubeflowV2DagRunnerConfig( default_image=configs.PIPELINE_IMAGE ) runner.KubeflowV2DagRunner( config=runner_config, output_filename=configs.PIPELINE_NAME + "_pipeline.json", ).run( kubeflow_pipeline.create_pipeline( pipeline_name=configs.PIPELINE_NAME, pipeline_root=configs.PIPELINE_ROOT, data_path=configs.DATA_PATH, schema_path=configs.SCHEMA_PATH, modules={ "training_fn": configs.TRAINING_FN, "preprocessing_fn": configs.PREPROCESSING_FN, }, eval_configs=configs.EVAL_CONFIGS, ai_platform_training_args=configs.GCP_AI_PLATFORM_TRAINING_ARGS, ai_platform_serving_args=configs.GCP_AI_PLATFORM_SERVING_ARGS, example_gen_beam_args=configs.EXAMPLE_GEN_BEAM_ARGS, transform_beam_args=configs.TRANSFORM_BEAM_ARGS, ) ) if __name__ == "__main__": logging.set_verbosity(logging.INFO) run()
deep-diver/mlops-hf-tf-vision-models
advanced_part1/local_runner.py
import os from absl import logging from tfx import v1 as tfx from pipeline import configs from pipeline import local_pipeline OUTPUT_DIR = "." PIPELINE_ROOT = os.path.join(OUTPUT_DIR, "tfx_pipeline_output", configs.PIPELINE_NAME) METADATA_PATH = os.path.join( OUTPUT_DIR, "tfx_metadata", configs.PIPELINE_NAME, "metadata.db" ) SERVING_MODEL_DIR = os.path.join(PIPELINE_ROOT, "serving_model") def run(): tfx.orchestration.LocalDagRunner().run( local_pipeline.create_pipeline( pipeline_name=configs.PIPELINE_NAME, pipeline_root=PIPELINE_ROOT, data_path=configs.DATA_PATH, schema_path=configs.SCHEMA_PATH, modules={ "training_fn": configs.TRAINING_FN, "preprocessing_fn": configs.PREPROCESSING_FN, }, eval_configs=configs.EVAL_CONFIGS, serving_model_dir=SERVING_MODEL_DIR, metadata_connection_config=tfx.orchestration.metadata.sqlite_metadata_connection_config( METADATA_PATH ), ) ) if __name__ == "__main__": logging.set_verbosity(logging.INFO) run()
deep-diver/mlops-hf-tf-vision-models
advanced_part2/kubeflow_runner.py
from absl import logging from tfx import v1 as tfx from tfx.orchestration.kubeflow.v2 import kubeflow_v2_dag_runner as runner from tfx.proto import tuner_pb2 from pipeline import configs from pipeline import kubeflow_pipeline def run(): runner_config = runner.KubeflowV2DagRunnerConfig( default_image=configs.PIPELINE_IMAGE ) runner.KubeflowV2DagRunner( config=runner_config, output_filename=configs.PIPELINE_NAME + "_pipeline.json", ).run( kubeflow_pipeline.create_pipeline( pipeline_name=configs.PIPELINE_NAME, pipeline_root=configs.PIPELINE_ROOT, data_path=configs.DATA_PATH, schema_path=configs.SCHEMA_PATH, modules={ "training_fn": configs.TRAINING_FN, "preprocessing_fn": configs.PREPROCESSING_FN, "tuner_fn": configs.TUNER_FN, }, eval_configs=configs.EVAL_CONFIGS, ai_platform_training_args=configs.GCP_AI_PLATFORM_TRAINING_ARGS, ai_platform_tuner_args=configs.GCP_AI_PLATFORM_TUNER_ARGS, tuner_args=tuner_pb2.TuneArgs( num_parallel_trials=configs.NUM_PARALLEL_TRIALS ), ai_platform_serving_args=configs.GCP_AI_PLATFORM_SERVING_ARGS, example_gen_beam_args=configs.EXAMPLE_GEN_BEAM_ARGS, transform_beam_args=configs.TRANSFORM_BEAM_ARGS, ) ) if __name__ == "__main__": logging.set_verbosity(logging.INFO) run()
deep-diver/mlops-hf-tf-vision-models
advanced_part2/local_runner.py
import os from absl import logging from tfx import v1 as tfx from pipeline import configs from pipeline import local_pipeline OUTPUT_DIR = "." PIPELINE_ROOT = os.path.join(OUTPUT_DIR, "tfx_pipeline_output", configs.PIPELINE_NAME) METADATA_PATH = os.path.join( OUTPUT_DIR, "tfx_metadata", configs.PIPELINE_NAME, "metadata.db" ) SERVING_MODEL_DIR = os.path.join(PIPELINE_ROOT, "serving_model") def run(): tfx.orchestration.LocalDagRunner().run( local_pipeline.create_pipeline( pipeline_name=configs.PIPELINE_NAME, pipeline_root=PIPELINE_ROOT, data_path=configs.DATA_PATH, schema_path=configs.SCHEMA_PATH, modules={ "training_fn": configs.TRAINING_FN, "preprocessing_fn": configs.PREPROCESSING_FN, "tuner_fn": configs.TUNER_FN, }, hyperparameters=configs.HYPER_PARAMETERS, eval_configs=configs.EVAL_CONFIGS, serving_model_dir=SERVING_MODEL_DIR, metadata_connection_config=tfx.orchestration.metadata.sqlite_metadata_connection_config( METADATA_PATH ), ) ) if __name__ == "__main__": logging.set_verbosity(logging.INFO) run()
deep-diver/mlops-hf-tf-vision-models
basic/kubeflow_runner.py
from absl import logging from tfx import v1 as tfx from tfx.orchestration.kubeflow.v2 import kubeflow_v2_dag_runner as runner from pipeline import configs from pipeline import kubeflow_pipeline def run(): runner_config = runner.KubeflowV2DagRunnerConfig( default_image=configs.PIPELINE_IMAGE ) runner.KubeflowV2DagRunner( config=runner_config, output_filename=configs.PIPELINE_NAME + "_pipeline.json", ).run( kubeflow_pipeline.create_pipeline( pipeline_name=configs.PIPELINE_NAME, pipeline_root=configs.PIPELINE_ROOT, data_path=configs.DATA_PATH, modules={ "training_fn": configs.TRAINING_FN, }, ai_platform_training_args=configs.GCP_AI_PLATFORM_TRAINING_ARGS, ai_platform_serving_args=configs.GCP_AI_PLATFORM_SERVING_ARGS, example_gen_beam_args=configs.EXAMPLE_GEN_BEAM_ARGS, ) ) if __name__ == "__main__": logging.set_verbosity(logging.INFO) run()
deep-diver/mlops-hf-tf-vision-models
basic/local_runner.py
import os from absl import logging from tfx import v1 as tfx from pipeline import configs from pipeline import local_pipeline OUTPUT_DIR = "." PIPELINE_ROOT = os.path.join(OUTPUT_DIR, "tfx_pipeline_output", configs.PIPELINE_NAME) METADATA_PATH = os.path.join( OUTPUT_DIR, "tfx_metadata", configs.PIPELINE_NAME, "metadata.db" ) SERVING_MODEL_DIR = os.path.join(PIPELINE_ROOT, "serving_model") def run(): tfx.orchestration.LocalDagRunner().run( local_pipeline.create_pipeline( pipeline_name=configs.PIPELINE_NAME, pipeline_root=PIPELINE_ROOT, data_path=configs.DATA_PATH, modules={ "training_fn": configs.TRAINING_FN, }, serving_model_dir=SERVING_MODEL_DIR, metadata_connection_config=tfx.orchestration.metadata.sqlite_metadata_connection_config( METADATA_PATH ), ) ) if __name__ == "__main__": logging.set_verbosity(logging.INFO) run()
deep-diver/mlops-hf-tf-vision-models
dataset/create_tfrecords.py
""" Script to generate TFRecord shards from the Sidewalks dataset as shown in this blog post: https://huggingface.co/blog/fine-tune-segformer. The recommended way to obtain TFRecord shards is via an Apache Beam Pipeline with an execution runner like Dataflow. Example: https://github.com/GoogleCloudPlatform/practical-ml-vision-book/blob/master/05_create_dataset/jpeg_to_tfrecord.py. Usage: python create_tfrecords --batch_size 16 python create_tfrecords --resize 256 # without --resize flag, no resizing is applied References: * https://github.com/GoogleCloudPlatform/practical-ml-vision-book/blob/master/05_create_dataset/05_split_tfrecord.ipynb * https://www.tensorflow.org/tutorials/images/segmentation """ import argparse import math import os from typing import Tuple import datasets import numpy as np import tensorflow as tf import tqdm from PIL import Image RESOLUTION = 256 def load_beans_dataset(args): hf_dataset_identifier = "beans" ds = datasets.load_dataset(hf_dataset_identifier) ds = ds.shuffle(seed=1) ds = ds["train"].train_test_split(test_size=args.split, seed=args.seed) train_ds = ds["train"] val_ds = ds["test"] return train_ds, val_ds def resize_img( image: tf.Tensor, label: tf.Tensor, resize: int ) -> Tuple[tf.Tensor, tf.Tensor]: image = tf.image.resize(image, (resize, resize)) return image, label def process_image( image: Image, label: Image, resize: int ) -> Tuple[tf.Tensor, tf.Tensor]: image = np.array(image) label = np.array(label) image = tf.convert_to_tensor(image) label = tf.convert_to_tensor(label) if resize: image, label = resize_img(image, label, resize) return image, label def _int64_feature(value): return tf.train.Feature(int64_list=tf.train.Int64List(value=value)) def _float_feature(value): return tf.train.Feature(float_list=tf.train.FloatList(value=value)) def create_tfrecord(image: Image, label: Image, resize: int): image, label = process_image(image, label, resize) image_dims = image.shape image = tf.reshape(image, [-1]) # flatten to 1D array label = tf.reshape(label, [-1]) # flatten to 1D array return tf.train.Example( features=tf.train.Features( feature={ "image": _float_feature(image.numpy()), "image_shape": _int64_feature( [image_dims[0], image_dims[1], image_dims[2]] ), "label": _int64_feature(label.numpy()), } ) ).SerializeToString() def write_tfrecords(root_dir, dataset, split, batch_size, resize): print(f"Preparing TFRecords for split: {split}.") for step in tqdm.tnrange(int(math.ceil(len(dataset) / batch_size))): temp_ds = dataset[step * batch_size : (step + 1) * batch_size] shard_size = len(temp_ds["image"]) filename = os.path.join( root_dir, "{}-{:02d}-{}.tfrec".format(split, step, shard_size) ) with tf.io.TFRecordWriter(filename) as out_file: for i in range(shard_size): image = temp_ds["image"][i] label = temp_ds["labels"][i] example = create_tfrecord(image, label, resize) out_file.write(example) print("Wrote file {} containing {} records".format(filename, shard_size)) def main(args): train_ds, val_ds = load_beans_dataset(args) print("Dataset loaded from HF.") if not os.path.exists(args.root_tfrecord_dir): os.makedirs(args.root_tfrecord_dir, exist_ok=True) print(args.resize) write_tfrecords( args.root_tfrecord_dir, train_ds, "train", args.batch_size, args.resize ) write_tfrecords(args.root_tfrecord_dir, val_ds, "val", args.batch_size, args.resize) def parse_args(): parser = argparse.ArgumentParser() parser.add_argument( "--split", help="Train and test split.", default=0.2, type=float ) parser.add_argument( "--seed", help="Seed to be used while performing train-test splits.", default=2022, type=int, ) parser.add_argument( "--root_tfrecord_dir", help="Root directory where the TFRecord shards will be serialized.", default="beans-tfrecords", type=str, ) parser.add_argument( "--batch_size", help="Number of samples to process in a batch before serializing a single TFRecord shard.", default=32, type=int, ) parser.add_argument( "--resize", help="Width and height size the image will be resized to. No resizing will be applied when this isn't set.", type=int, ) args = parser.parse_args() return args if __name__ == "__main__": args = parse_args() main(args)
deep-diver/mlops-hf-tf-vision-models
hf_integration/kubeflow_runner.py
from absl import logging from tfx import v1 as tfx from tfx.orchestration.kubeflow.v2 import kubeflow_v2_dag_runner as runner from tfx.proto import tuner_pb2 from pipeline import configs from pipeline import kubeflow_pipeline def run(): runner_config = runner.KubeflowV2DagRunnerConfig( default_image=configs.PIPELINE_IMAGE ) runner.KubeflowV2DagRunner( config=runner_config, output_filename=configs.PIPELINE_NAME + "_pipeline.json", ).run( kubeflow_pipeline.create_pipeline( pipeline_name=configs.PIPELINE_NAME, pipeline_root=configs.PIPELINE_ROOT, data_path=configs.DATA_PATH, schema_path=configs.SCHEMA_PATH, modules={ "training_fn": configs.TRAINING_FN, "preprocessing_fn": configs.PREPROCESSING_FN, "tuner_fn": configs.TUNER_FN, }, eval_configs=configs.EVAL_CONFIGS, ai_platform_training_args=configs.GCP_AI_PLATFORM_TRAINING_ARGS, ai_platform_tuner_args=configs.GCP_AI_PLATFORM_TUNER_ARGS, tuner_args=tuner_pb2.TuneArgs( num_parallel_trials=configs.NUM_PARALLEL_TRIALS ), ai_platform_serving_args=configs.GCP_AI_PLATFORM_SERVING_ARGS, example_gen_beam_args=configs.EXAMPLE_GEN_BEAM_ARGS, transform_beam_args=configs.TRANSFORM_BEAM_ARGS, hf_pusher_args=configs.HF_PUSHER_ARGS, ) ) if __name__ == "__main__": logging.set_verbosity(logging.INFO) run()
deep-diver/mlops-hf-tf-vision-models
hf_integration/local_runner.py
import os from absl import logging from tfx import v1 as tfx from pipeline import configs from pipeline import local_pipeline OUTPUT_DIR = "." PIPELINE_ROOT = os.path.join(OUTPUT_DIR, "tfx_pipeline_output", configs.PIPELINE_NAME) METADATA_PATH = os.path.join( OUTPUT_DIR, "tfx_metadata", configs.PIPELINE_NAME, "metadata.db" ) SERVING_MODEL_DIR = os.path.join(PIPELINE_ROOT, "serving_model") def run(): tfx.orchestration.LocalDagRunner().run( local_pipeline.create_pipeline( pipeline_name=configs.PIPELINE_NAME, pipeline_root=PIPELINE_ROOT, data_path=configs.DATA_PATH, schema_path=configs.SCHEMA_PATH, modules={ "training_fn": configs.TRAINING_FN, "preprocessing_fn": configs.PREPROCESSING_FN, "tuner_fn": configs.TUNER_FN, }, hyperparameters=configs.HYPER_PARAMETERS, eval_configs=configs.EVAL_CONFIGS, serving_model_dir=SERVING_MODEL_DIR, metadata_connection_config=tfx.orchestration.metadata.sqlite_metadata_connection_config( METADATA_PATH ), ) ) if __name__ == "__main__": logging.set_verbosity(logging.INFO) run()
deep-diver/mlops-hf-tf-vision-models
intermediate/kubeflow_runner.py
from absl import logging from tfx import v1 as tfx from tfx.orchestration.kubeflow.v2 import kubeflow_v2_dag_runner as runner from pipeline import configs from pipeline import kubeflow_pipeline def run(): runner_config = runner.KubeflowV2DagRunnerConfig( default_image=configs.PIPELINE_IMAGE ) runner.KubeflowV2DagRunner( config=runner_config, output_filename=configs.PIPELINE_NAME + "_pipeline.json", ).run( kubeflow_pipeline.create_pipeline( pipeline_name=configs.PIPELINE_NAME, pipeline_root=configs.PIPELINE_ROOT, data_path=configs.DATA_PATH, schema_path=configs.SCHEMA_PATH, modules={ "training_fn": configs.TRAINING_FN, "preprocessing_fn": configs.PREPROCESSING_FN, }, ai_platform_training_args=configs.GCP_AI_PLATFORM_TRAINING_ARGS, ai_platform_serving_args=configs.GCP_AI_PLATFORM_SERVING_ARGS, example_gen_beam_args=configs.EXAMPLE_GEN_BEAM_ARGS, transform_beam_args=configs.TRANSFORM_BEAM_ARGS, ) ) if __name__ == "__main__": logging.set_verbosity(logging.INFO) run()
deep-diver/mlops-hf-tf-vision-models
intermediate/local_runner.py
import os from absl import logging from tfx import v1 as tfx from pipeline import configs from pipeline import local_pipeline OUTPUT_DIR = "." PIPELINE_ROOT = os.path.join(OUTPUT_DIR, "tfx_pipeline_output", configs.PIPELINE_NAME) METADATA_PATH = os.path.join( OUTPUT_DIR, "tfx_metadata", configs.PIPELINE_NAME, "metadata.db" ) SERVING_MODEL_DIR = os.path.join(PIPELINE_ROOT, "serving_model") def run(): tfx.orchestration.LocalDagRunner().run( local_pipeline.create_pipeline( pipeline_name=configs.PIPELINE_NAME, pipeline_root=PIPELINE_ROOT, data_path=configs.DATA_PATH, schema_path=configs.SCHEMA_PATH, modules={ "training_fn": configs.TRAINING_FN, "preprocessing_fn": configs.PREPROCESSING_FN, }, serving_model_dir=SERVING_MODEL_DIR, metadata_connection_config=tfx.orchestration.metadata.sqlite_metadata_connection_config( METADATA_PATH ), ) ) if __name__ == "__main__": logging.set_verbosity(logging.INFO) run()
deep-diver/mlops-hf-tf-vision-models
notebooks/advanced_part1.ipynb
data_path = "gs://beans-lowres/tfrecords" local_data_path = "data" model_file = "modules/model.py" model_fn = "modules.model.run_fn" proprocessing_file = "modules/preprocessing.py" preprocessing_fn = "modules.preprocessing.preprocessing_fn" schema_file = "schema.pbtxt"import tfx tfx.__version__from tfx import v1 as tfx from tfx.orchestration.experimental.interactive.interactive_context import InteractiveContext from tfx.components import ImportExampleGen from tfx.components import StatisticsGen from tfx.components import SchemaGen from tfx.components import ExampleValidator from tfx.components import Transform from tfx.components import Trainer from tfx.components import Evaluator from tfx.components import Pusher from tfx.proto import example_gen_pb2 from tfx.types import Channel from tfx.types.standard_artifacts import Model from tfx.types.standard_artifacts import ModelBlessing from tfx.dsl.components.common import resolver from tfx.dsl.experimental.latest_blessed_model_resolver import LatestBlessedModelResolver import tensorflow_model_analysis as tfmacontext = InteractiveContext()input_config = example_gen_pb2.Input( splits=[ example_gen_pb2.Input.Split(name="train", pattern="train-*.tfrec"), example_gen_pb2.Input.Split(name="eval", pattern="val-*.tfrec"), ] ) example_gen = ImportExampleGen( input_base=local_data_path, input_config=input_config ) context.run(example_gen)statistics_gen = StatisticsGen(examples=example_gen.outputs['examples']) context.run(statistics_gen)context.show(statistics_gen.outputs['statistics'])%%writefile {schema_file} feature { name: "image" type: FLOAT presence { min_fraction: 1.0 } float_domain { min: 0 max: 255 } shape { dim { size: 256 } dim { size: 256 } dim { size: 3 } } } feature { name: "image_shape" type: INT presence { min_fraction: 1.0 } shape { dim { size: 3 } } } feature { name: "label" type: INT presence { min_fraction: 1.0 } int_domain { min: 0 max: 2 } shape { dim { size: 1 } } }schema_gen = tfx.components.ImportSchemaGen( schema_file=schema_file) context.run(schema_gen)example_validator = ExampleValidator( statistics=statistics_gen.outputs['statistics'], schema=schema_gen.outputs['schema'] ) context.run(example_validator)context.show(example_validator.outputs['anomalies'])%%writefile {proprocessing_file} import tensorflow as tf IMAGE_KEY = "image" LABEL_KEY = "label" MODEL_INPUT_IMAGE_KEY = "pixel_values" MODEL_INPUT_LABEL_KEY = "labels" INPUT_IMG_SIZE = 224 def preprocessing_fn(inputs): """tf.transform's callback function for preprocessing inputs. Args: inputs: map from feature keys to raw not-yet-transformed features. Returns: Map from string feature key to transformed feature operations. """ # print(inputs) outputs = {} inputs[IMAGE_KEY] = tf.image.resize( inputs[IMAGE_KEY], [INPUT_IMG_SIZE, INPUT_IMG_SIZE] ) inputs[IMAGE_KEY] = inputs[IMAGE_KEY] / 255.0 inputs[IMAGE_KEY] = tf.transpose(inputs[IMAGE_KEY], [0, 3, 1, 2]) outputs[MODEL_INPUT_IMAGE_KEY] = inputs[IMAGE_KEY] outputs[MODEL_INPUT_LABEL_KEY] = inputs[LABEL_KEY] return outputstransform = Transform( examples=example_gen.outputs['examples'], schema=schema_gen.outputs['schema'], preprocessing_fn=preprocessing_fn)context.run(transform)%%writefile {model_file} from typing import List, Dict, Tuple import absl import tensorflow as tf import tensorflow_transform as tft from transformers import ViTFeatureExtractor, TFViTForImageClassification from tfx.components.trainer.fn_args_utils import FnArgs from tfx_bsl.tfxio import dataset_options from tfx.components.trainer.fn_args_utils import DataAccessor feature_extractor = ViTFeatureExtractor() _TRAIN_LENGTH = 128 _EVAL_LENGTH = 128 _TRAIN_BATCH_SIZE = 8 _EVAL_BATCH_SIZE = 8 _EPOCHS = 1 _LABELS = ['angular_leaf_spot', 'bean_rust', 'healthy'] _CONCRETE_INPUT = "pixel_values" _MODEL_INPUT_LABEL_KEY = "labels" def INFO(text: str): absl.logging.info(text) def _normalize_img( img, mean=feature_extractor.image_mean, std=feature_extractor.image_std ): img = img / 255 mean = tf.constant(mean) std = tf.constant(std) return (img - mean) / std def _preprocess_serving(string_input): decoded_input = tf.io.decode_base64(string_input) decoded = tf.io.decode_jpeg(decoded_input, channels=3) resized = tf.image.resize(decoded, size=(224, 224)) normalized = _normalize_img(resized) normalized = tf.transpose( normalized, (2, 0, 1) ) # Since HF models are channel-first. return normalized @tf.function(input_signature=[tf.TensorSpec([None], tf.string)]) def _preprocess_fn(string_input): decoded_images = tf.map_fn( _preprocess_serving, string_input, dtype=tf.float32, back_prop=False ) return {_CONCRETE_INPUT: decoded_images} def _model_exporter(model: tf.keras.Model): m_call = tf.function(model.call).get_concrete_function( tf.TensorSpec( shape=[None, 3, 224, 224], dtype=tf.float32, name=_CONCRETE_INPUT ) ) @tf.function(input_signature=[tf.TensorSpec([None], tf.string)]) def serving_fn(string_input): labels = tf.constant(list(model.config.id2label.values()), dtype=tf.string) images = _preprocess_fn(string_input) predictions = m_call(**images) indices = tf.argmax(predictions.logits, axis=1) pred_source = tf.gather(params=labels, indices=indices) probs = tf.nn.softmax(predictions.logits, axis=1) pred_confidence = tf.reduce_max(probs, axis=1) return {"label": pred_source, "confidence": pred_confidence} return serving_fn ### def _transform_features_signature( model: tf.keras.Model, tf_transform_output: tft.TFTransformOutput ): """ transform_features_signature simply returns a function that transforms any data of the type of tf.Example which is denoted as the type of sta ndard_artifacts.Examples in TFX. The purpose of this function is to ap ply Transform Graph obtained from Transform component to the data prod uced by ImportExampleGen. This function will be used in the Evaluator component, so the raw evaluation inputs from ImportExampleGen can be a pporiately transformed that the model could understand. """ # basically, what Transform component emits is a SavedModel that knows # how to transform data. transform_features_layer() simply returns the # layer from the Transform. model.tft_layer = tf_transform_output.transform_features_layer() @tf.function( input_signature=[tf.TensorSpec(shape=[None], dtype=tf.string, name="examples")] ) def serve_tf_examples_fn(serialized_tf_examples): """ raw_feature_spec returns a set of feature maps(dict) for the input TFRecords based on the knowledge that Transform component has lear ned(learn doesn't mean training here). By using this information, the raw data from ImportExampleGen could be parsed with tf.io.parse _example utility function. Then, it is passed to the model.tft_layer, so the final output we get is the transformed data of the raw input. """ feature_spec = tf_transform_output.raw_feature_spec() parsed_features = tf.io.parse_example(serialized_tf_examples, feature_spec) transformed_features = model.tft_layer(parsed_features) return transformed_features return serve_tf_examples_fn def _tf_examples_serving_signature(model, tf_transform_output): """ tf_examples_serving_signature simply returns a function that performs data transformation(preprocessing) and model prediction in a sequential manner. How data transformation is done is idential to the process of transform_features_signature function. """ model.tft_layer = tf_transform_output.transform_features_layer() @tf.function( input_signature=[tf.TensorSpec(shape=[None], dtype=tf.string, name="examples")] ) def serve_tf_examples_fn( serialized_tf_example: tf.Tensor, ) -> Dict[str, tf.Tensor]: raw_feature_spec = tf_transform_output.raw_feature_spec() raw_features = tf.io.parse_example(serialized_tf_example, raw_feature_spec) transformed_features = model.tft_layer(raw_features) logits = model(transformed_features).logits return {_MODEL_INPUT_LABEL_KEY: logits} return serve_tf_examples_fn ### def _input_fn( file_pattern: List[str], data_accessor: DataAccessor, tf_transform_output: tft.TFTransformOutput, is_train: bool = False, batch_size: int = 32, ) -> tf.data.Dataset: INFO(f"Reading data from: {file_pattern}") dataset = data_accessor.tf_dataset_factory( file_pattern, dataset_options.TensorFlowDatasetOptions( batch_size=batch_size, label_key=_MODEL_INPUT_LABEL_KEY ), tf_transform_output.transformed_metadata.schema, ) return dataset def _build_model(): id2label={str(i): c for i, c in enumerate(_LABELS)} label2id={c: str(i) for i, c in enumerate(_LABELS)} model = TFViTForImageClassification.from_pretrained( "google/vit-base-patch16-224-in21k", num_labels=len(_LABELS), label2id=label2id, id2label=id2label, ) model.layers[0].trainable=False loss = tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True) model.compile(optimizer='adam', loss=loss, metrics=["accuracy"]) return model def run_fn(fn_args: FnArgs): tf_transform_output = tft.TFTransformOutput(fn_args.transform_output) train_dataset = _input_fn( fn_args.train_files, fn_args.data_accessor, tf_transform_output, is_train=True, batch_size=_TRAIN_BATCH_SIZE, ) eval_dataset = _input_fn( fn_args.eval_files, fn_args.data_accessor, tf_transform_output, is_train=False, batch_size=_EVAL_BATCH_SIZE, ) model = _build_model() model.fit( train_dataset, steps_per_epoch=_TRAIN_LENGTH // _TRAIN_BATCH_SIZE, validation_data=eval_dataset, validation_steps=_EVAL_LENGTH // _EVAL_BATCH_SIZE, epochs=_EPOCHS, ) model.save( fn_args.serving_model_dir, save_format="tf", signatures={ "serving_default": _model_exporter(model), "transform_features": _transform_features_signature( model, tf_transform_output ), "from_examples": _tf_examples_serving_signature(model, tf_transform_output), }, )trainer = Trainer( run_fn=model_fn, transformed_examples=transform.outputs["transformed_examples"], transform_graph=transform.outputs["transform_graph"], schema=schema_gen.outputs["schema"], )context.run(trainer)model_resolver = resolver.Resolver( strategy_class=LatestBlessedModelResolver, model=Channel(type=Model), model_blessing=Channel(type=ModelBlessing), ).with_id("latest_blessed_model_resolver")context.run(model_resolver)eval_configs = tfma.EvalConfig( model_specs=[ tfma.ModelSpec( signature_name="from_examples", preprocessing_function_names=["transform_features"], label_key="labels", prediction_key="labels", ) ], slicing_specs=[tfma.SlicingSpec()], metrics_specs=[ tfma.MetricsSpec( metrics=[ tfma.MetricConfig( class_name="SparseCategoricalAccuracy", threshold=tfma.MetricThreshold( value_threshold=tfma.GenericValueThreshold( lower_bound={"value": 0.55} ), # Change threshold will be ignored if there is no # baseline model resolved from MLMD (first run). change_threshold=tfma.GenericChangeThreshold( direction=tfma.MetricDirection.HIGHER_IS_BETTER, absolute={"value": -1e-3}, ), ), ) ] ) ], ) evaluator = Evaluator( examples=example_gen.outputs["examples"], model=trainer.outputs["model"], baseline_model=model_resolver.outputs["model"], eval_config=eval_configs, )context.run(evaluator)
deep-diver/mlops-hf-tf-vision-models
notebooks/advanced_part2.ipynb
data_path = "gs://beans-lowres/tfrecords" local_data_path = "data" model_file = "modules/model.py" model_fn = "modules.model.run_fn" tuner_fn = "modules.model.tuner_fn" proprocessing_file = "modules/preprocessing.py" preprocessing_fn = "modules.preprocessing.preprocessing_fn" schema_file = "schema.pbtxt"import tfx tfx.__version__from tfx import v1 as tfx from tfx.orchestration.experimental.interactive.interactive_context import InteractiveContext from tfx.components import ImportExampleGen from tfx.components import StatisticsGen from tfx.components import SchemaGen from tfx.components import ExampleValidator from tfx.components import Transform from tfx.components import Trainer from tfx.components import Tuner from tfx.components import Evaluator from tfx.components import Pusher from tfx.proto import example_gen_pb2 from tfx.types import Channel from tfx.types.standard_artifacts import Model from tfx.types.standard_artifacts import ModelBlessing from tfx.dsl.components.common import resolver from tfx.dsl.experimental.latest_blessed_model_resolver import LatestBlessedModelResolver import tensorflow_model_analysis as tfmacontext = InteractiveContext()input_config = example_gen_pb2.Input( splits=[ example_gen_pb2.Input.Split(name="train", pattern="train-*.tfrec"), example_gen_pb2.Input.Split(name="eval", pattern="val-*.tfrec"), ] ) example_gen = ImportExampleGen( input_base=local_data_path, input_config=input_config ) context.run(example_gen)statistics_gen = StatisticsGen(examples=example_gen.outputs['examples']) context.run(statistics_gen)context.show(statistics_gen.outputs['statistics'])%%writefile {schema_file} feature { name: "image" type: FLOAT presence { min_fraction: 1.0 } float_domain { min: 0 max: 255 } shape { dim { size: 256 } dim { size: 256 } dim { size: 3 } } } feature { name: "image_shape" type: INT presence { min_fraction: 1.0 } shape { dim { size: 3 } } } feature { name: "label" type: INT presence { min_fraction: 1.0 } int_domain { min: 0 max: 2 } shape { dim { size: 1 } } }schema_gen = tfx.components.ImportSchemaGen( schema_file=schema_file) context.run(schema_gen)example_validator = ExampleValidator( statistics=statistics_gen.outputs['statistics'], schema=schema_gen.outputs['schema'] ) context.run(example_validator)context.show(example_validator.outputs['anomalies'])%%writefile {proprocessing_file} import tensorflow as tf IMAGE_KEY = "image" LABEL_KEY = "label" MODEL_INPUT_IMAGE_KEY = "pixel_values" MODEL_INPUT_LABEL_KEY = "labels" INPUT_IMG_SIZE = 224 def preprocessing_fn(inputs): """tf.transform's callback function for preprocessing inputs. Args: inputs: map from feature keys to raw not-yet-transformed features. Returns: Map from string feature key to transformed feature operations. """ # print(inputs) outputs = {} inputs[IMAGE_KEY] = tf.image.resize( inputs[IMAGE_KEY], [INPUT_IMG_SIZE, INPUT_IMG_SIZE] ) inputs[IMAGE_KEY] = inputs[IMAGE_KEY] / 255.0 inputs[IMAGE_KEY] = tf.transpose(inputs[IMAGE_KEY], [0, 3, 1, 2]) outputs[MODEL_INPUT_IMAGE_KEY] = inputs[IMAGE_KEY] outputs[MODEL_INPUT_LABEL_KEY] = inputs[LABEL_KEY] return outputstransform = Transform( examples=example_gen.outputs['examples'], schema=schema_gen.outputs['schema'], preprocessing_fn=preprocessing_fn)context.run(transform)%%writefile {model_file} from typing import List, Dict, Tuple import absl import tensorflow as tf import keras_tuner import tensorflow_transform as tft from transformers import ViTFeatureExtractor, TFViTForImageClassification from tfx.components.trainer.fn_args_utils import FnArgs from tfx.v1.components import TunerFnResult from tfx_bsl.tfxio import dataset_options from tfx.components.trainer.fn_args_utils import DataAccessor feature_extractor = ViTFeatureExtractor() _TRAIN_LENGTH = 128 _EVAL_LENGTH = 128 _TRAIN_BATCH_SIZE = 8 _EVAL_BATCH_SIZE = 8 _EPOCHS = 1 _LABELS = ['angular_leaf_spot', 'bean_rust', 'healthy'] _CONCRETE_INPUT = "pixel_values" _MODEL_INPUT_LABEL_KEY = "labels" def INFO(text: str): absl.logging.info(text) def _normalize_img( img, mean=feature_extractor.image_mean, std=feature_extractor.image_std ): img = img / 255 mean = tf.constant(mean) std = tf.constant(std) return (img - mean) / std def _preprocess_serving(string_input): decoded_input = tf.io.decode_base64(string_input) decoded = tf.io.decode_jpeg(decoded_input, channels=3) resized = tf.image.resize(decoded, size=(224, 224)) normalized = _normalize_img(resized) normalized = tf.transpose( normalized, (2, 0, 1) ) # Since HF models are channel-first. return normalized @tf.function(input_signature=[tf.TensorSpec([None], tf.string)]) def _preprocess_fn(string_input): decoded_images = tf.map_fn( _preprocess_serving, string_input, dtype=tf.float32, back_prop=False ) return {_CONCRETE_INPUT: decoded_images} def _model_exporter(model: tf.keras.Model): m_call = tf.function(model.call).get_concrete_function( tf.TensorSpec( shape=[None, 3, 224, 224], dtype=tf.float32, name=_CONCRETE_INPUT ) ) @tf.function(input_signature=[tf.TensorSpec([None], tf.string)]) def serving_fn(string_input): labels = tf.constant(list(model.config.id2label.values()), dtype=tf.string) images = _preprocess_fn(string_input) predictions = m_call(**images) indices = tf.argmax(predictions.logits, axis=1) pred_source = tf.gather(params=labels, indices=indices) probs = tf.nn.softmax(predictions.logits, axis=1) pred_confidence = tf.reduce_max(probs, axis=1) return {"label": pred_source, "confidence": pred_confidence} return serving_fn ### def _transform_features_signature( model: tf.keras.Model, tf_transform_output: tft.TFTransformOutput ): """ transform_features_signature simply returns a function that transforms any data of the type of tf.Example which is denoted as the type of sta ndard_artifacts.Examples in TFX. The purpose of this function is to ap ply Transform Graph obtained from Transform component to the data prod uced by ImportExampleGen. This function will be used in the Evaluator component, so the raw evaluation inputs from ImportExampleGen can be a pporiately transformed that the model could understand. """ # basically, what Transform component emits is a SavedModel that knows # how to transform data. transform_features_layer() simply returns the # layer from the Transform. model.tft_layer = tf_transform_output.transform_features_layer() @tf.function( input_signature=[tf.TensorSpec(shape=[None], dtype=tf.string, name="examples")] ) def serve_tf_examples_fn(serialized_tf_examples): """ raw_feature_spec returns a set of feature maps(dict) for the input TFRecords based on the knowledge that Transform component has lear ned(learn doesn't mean training here). By using this information, the raw data from ImportExampleGen could be parsed with tf.io.parse _example utility function. Then, it is passed to the model.tft_layer, so the final output we get is the transformed data of the raw input. """ feature_spec = tf_transform_output.raw_feature_spec() parsed_features = tf.io.parse_example(serialized_tf_examples, feature_spec) transformed_features = model.tft_layer(parsed_features) return transformed_features return serve_tf_examples_fn def _tf_examples_serving_signature(model, tf_transform_output): """ tf_examples_serving_signature simply returns a function that performs data transformation(preprocessing) and model prediction in a sequential manner. How data transformation is done is idential to the process of transform_features_signature function. """ model.tft_layer = tf_transform_output.transform_features_layer() @tf.function( input_signature=[tf.TensorSpec(shape=[None], dtype=tf.string, name="examples")] ) def serve_tf_examples_fn( serialized_tf_example: tf.Tensor, ) -> Dict[str, tf.Tensor]: raw_feature_spec = tf_transform_output.raw_feature_spec() raw_features = tf.io.parse_example(serialized_tf_example, raw_feature_spec) transformed_features = model.tft_layer(raw_features) logits = model(transformed_features).logits return {_MODEL_INPUT_LABEL_KEY: logits} return serve_tf_examples_fn ### def _input_fn( file_pattern: List[str], data_accessor: DataAccessor, tf_transform_output: tft.TFTransformOutput, is_train: bool = False, batch_size: int = 32, ) -> tf.data.Dataset: INFO(f"Reading data from: {file_pattern}") dataset = data_accessor.tf_dataset_factory( file_pattern, dataset_options.TensorFlowDatasetOptions( batch_size=batch_size, label_key=_MODEL_INPUT_LABEL_KEY ), tf_transform_output.transformed_metadata.schema, ) return dataset def _get_hyperparameters() -> keras_tuner.HyperParameters: hp = keras_tuner.HyperParameters() hp.Choice("learning_rate", [1e-3, 1e-2], default=1e-3) return hp def _build_model(hparams: keras_tuner.HyperParameters): id2label={str(i): c for i, c in enumerate(_LABELS)} label2id={c: str(i) for i, c in enumerate(_LABELS)} model = TFViTForImageClassification.from_pretrained( "google/vit-base-patch16-224-in21k", num_labels=len(_LABELS), label2id=label2id, id2label=id2label, ) model.layers[0].trainable=False loss = tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True) optimizer = tf.keras.optimizers.Adam(learning_rate=hparams.get("learning_rate")) model.compile(optimizer=optimizer, loss=loss, metrics=["accuracy"]) return model def run_fn(fn_args: FnArgs): tf_transform_output = tft.TFTransformOutput(fn_args.transform_output) train_dataset = _input_fn( fn_args.train_files, fn_args.data_accessor, tf_transform_output, is_train=True, batch_size=_TRAIN_BATCH_SIZE, ) eval_dataset = _input_fn( fn_args.eval_files, fn_args.data_accessor, tf_transform_output, is_train=False, batch_size=_EVAL_BATCH_SIZE, ) hparams = keras_tuner.HyperParameters.from_config(fn_args.hyperparameters) INFO(f"HyperParameters for training: {hparams.get_config()}") model = _build_model(hparams) model.fit( train_dataset, steps_per_epoch=_TRAIN_LENGTH // _TRAIN_BATCH_SIZE, validation_data=eval_dataset, validation_steps=_EVAL_LENGTH // _EVAL_BATCH_SIZE, epochs=_EPOCHS, ) model.save( fn_args.serving_model_dir, save_format="tf", signatures={ "serving_default": _model_exporter(model), "transform_features": _transform_features_signature( model, tf_transform_output ), "from_examples": _tf_examples_serving_signature(model, tf_transform_output), }, ) def tuner_fn(fn_args: FnArgs) -> TunerFnResult: tuner = keras_tuner.RandomSearch( _build_model, max_trials=6, hyperparameters=_get_hyperparameters(), allow_new_entries=False, objective=keras_tuner.Objective("val_accuracy", "max"), directory=fn_args.working_dir, project_name="img_classification_tuning", ) tf_transform_output = tft.TFTransformOutput(fn_args.transform_graph_path) train_dataset = _input_fn( fn_args.train_files, fn_args.data_accessor, tf_transform_output, is_train=True, batch_size=_TRAIN_BATCH_SIZE, ) eval_dataset = _input_fn( fn_args.eval_files, fn_args.data_accessor, tf_transform_output, is_train=False, batch_size=_EVAL_BATCH_SIZE, ) return TunerFnResult( tuner=tuner, fit_kwargs={ "x": train_dataset, "validation_data": eval_dataset, "steps_per_epoch": _TRAIN_LENGTH // _TRAIN_BATCH_SIZE, "validation_steps": _EVAL_LENGTH // _EVAL_BATCH_SIZE, }, ) tuner = Tuner( tuner_fn=tuner_fn, examples=transform.outputs["transformed_examples"], schema=schema_gen.outputs["schema"], transform_graph=transform.outputs["transform_graph"], )context.run(tuner)trainer = Trainer( run_fn=model_fn, transformed_examples=transform.outputs["transformed_examples"], transform_graph=transform.outputs["transform_graph"], schema=schema_gen.outputs["schema"], hyperparameters=tuner.outputs["best_hyperparameters"], )context.run(trainer)model_resolver = resolver.Resolver( strategy_class=LatestBlessedModelResolver, model=Channel(type=Model), model_blessing=Channel(type=ModelBlessing), ).with_id("latest_blessed_model_resolver")context.run(model_resolver)eval_configs = tfma.EvalConfig( model_specs=[ tfma.ModelSpec( signature_name="from_examples", preprocessing_function_names=["transform_features"], label_key="labels", prediction_key="labels", ) ], slicing_specs=[tfma.SlicingSpec()], metrics_specs=[ tfma.MetricsSpec( metrics=[ tfma.MetricConfig( class_name="SparseCategoricalAccuracy", threshold=tfma.MetricThreshold( value_threshold=tfma.GenericValueThreshold( lower_bound={"value": 0.55} ), # Change threshold will be ignored if there is no # baseline model resolved from MLMD (first run). change_threshold=tfma.GenericChangeThreshold( direction=tfma.MetricDirection.HIGHER_IS_BETTER, absolute={"value": -1e-3}, ), ), ) ] ) ], ) evaluator = Evaluator( examples=example_gen.outputs["examples"], model=trainer.outputs["model"], baseline_model=model_resolver.outputs["model"], eval_config=eval_configs, )context.run(evaluator)
deep-diver/mlops-hf-tf-vision-models
notebooks/basic.ipynb
data_path = "gs://beans-lowres/tfrecords" local_data_path = "data" model_file = "modules/model.py" model_fn = "modules.model.run_fn"import tfx tfx.__version__from tfx import v1 as tfx from tfx.orchestration.experimental.interactive.interactive_context import InteractiveContext from tfx.components import ImportExampleGen from tfx.components import Trainer from tfx.components import Pusher from tfx.components import StatisticsGen from tfx.proto import example_gen_pb2context = InteractiveContext()input_config = example_gen_pb2.Input( splits=[ example_gen_pb2.Input.Split(name="train", pattern="train-*.tfrec"), example_gen_pb2.Input.Split(name="eval", pattern="val-*.tfrec"), ] ) example_gen = ImportExampleGen( input_base=local_data_path, input_config=input_config )context.run(example_gen)%%writefile {model_file} from typing import List, Dict, Tuple import absl import tensorflow as tf from transformers import ViTFeatureExtractor, TFViTForImageClassification from tfx.components.trainer.fn_args_utils import FnArgs feature_extractor = ViTFeatureExtractor() _TRAIN_LENGTH = 128 _EVAL_LENGTH = 128 _TRAIN_BATCH_SIZE = 8 _EVAL_BATCH_SIZE = 8 _EPOCHS = 1 _LABELS = ['angular_leaf_spot', 'bean_rust', 'healthy'] _CONCRETE_INPUT = "pixel_values" def INFO(text: str): absl.logging.info(text) ### def _normalize_img( img, mean=feature_extractor.image_mean, std=feature_extractor.image_std ): img = img / 255 mean = tf.constant(mean) std = tf.constant(std) return (img - mean) / std def _preprocess_serving(string_input): decoded_input = tf.io.decode_base64(string_input) decoded = tf.io.decode_jpeg(decoded_input, channels=3) resized = tf.image.resize(decoded, size=(224, 224)) normalized = _normalize_img(resized) normalized = tf.transpose( normalized, (2, 0, 1) ) # Since HF models are channel-first. return normalized @tf.function(input_signature=[tf.TensorSpec([None], tf.string)]) def _preprocess_fn(string_input): decoded_images = tf.map_fn( _preprocess_serving, string_input, dtype=tf.float32, back_prop=False ) return {_CONCRETE_INPUT: decoded_images} def _model_exporter(model: tf.keras.Model): m_call = tf.function(model.call).get_concrete_function( tf.TensorSpec( shape=[None, 3, 224, 224], dtype=tf.float32, name=_CONCRETE_INPUT ) ) @tf.function(input_signature=[tf.TensorSpec([None], tf.string)]) def serving_fn(string_input): labels = tf.constant(list(model.config.id2label.values()), dtype=tf.string) images = _preprocess_fn(string_input) predictions = m_call(**images) indices = tf.argmax(predictions.logits, axis=1) pred_source = tf.gather(params=labels, indices=indices) probs = tf.nn.softmax(predictions.logits, axis=1) pred_confidence = tf.reduce_max(probs, axis=1) return {"label": pred_source, "confidence": pred_confidence} return serving_fn ### def _parse_tfr(proto): feature_description = { "image": tf.io.VarLenFeature(tf.float32), "image_shape": tf.io.VarLenFeature(tf.int64), "label": tf.io.VarLenFeature(tf.int64), } rec = tf.io.parse_single_example(proto, feature_description) image_shape = tf.sparse.to_dense(rec["image_shape"]) image = tf.reshape(tf.sparse.to_dense(rec["image"]), image_shape) label = tf.sparse.to_dense(rec["label"]) return {"pixel_values": image, "labels": label} def _preprocess(example_batch): images = example_batch["pixel_values"] images = tf.transpose(images, perm=[0, 1, 2, 3]) # (batch_size, height, width, num_channels) images = tf.image.resize(images, (224, 224)) images = tf.transpose(images, perm=[0, 3, 1, 2]) labels = example_batch["labels"] labels = tf.transpose(labels, perm=[0, 1]) # So, that TF can evaluation the shapes. return {"pixel_values": images, "labels": labels} def _input_fn( file_pattern: List[str], batch_size: int = 32, is_train: bool = False, ) -> tf.data.Dataset: INFO(f"Reading data from: {file_pattern}") dataset = tf.data.TFRecordDataset( tf.io.gfile.glob(file_pattern[0] + ".gz"), num_parallel_reads=tf.data.AUTOTUNE, compression_type="GZIP", ).map(_parse_tfr, num_parallel_calls=tf.data.AUTOTUNE) if is_train: dataset = dataset.shuffle(batch_size * 2) dataset = dataset.batch(batch_size) dataset = dataset.prefetch(tf.data.AUTOTUNE) dataset = dataset.map(_preprocess) return dataset def _build_model(): id2label={str(i): c for i, c in enumerate(_LABELS)} label2id={c: str(i) for i, c in enumerate(_LABELS)} model = TFViTForImageClassification.from_pretrained( "google/vit-base-patch16-224-in21k", num_labels=len(_LABELS), label2id=label2id, id2label=id2label, ) loss = tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True) model.compile(optimizer='adam', loss=loss, metrics=["accuracy"]) return model def run_fn(fn_args: FnArgs): train_dataset = _input_fn( fn_args.train_files, is_train=True, batch_size=_TRAIN_BATCH_SIZE, ) eval_dataset = _input_fn( fn_args.eval_files, is_train=False, batch_size=_EVAL_BATCH_SIZE, ) model = _build_model() model.fit( train_dataset, steps_per_epoch=_TRAIN_LENGTH // _TRAIN_BATCH_SIZE, validation_data=eval_dataset, validation_steps=_EVAL_LENGTH // _TRAIN_BATCH_SIZE, epochs=_EPOCHS, ) model.save( fn_args.serving_model_dir, save_format="tf", signatures=_model_exporter(model) )trainer = Trainer( run_fn=model_fn, examples=example_gen.outputs["examples"], )context.run(trainer)
deep-diver/mlops-hf-tf-vision-models
notebooks/intermediate.ipynb
data_path = "gs://beans-lowres/tfrecords" local_data_path = "data" model_file = "modules/model.py" model_fn = "modules.model.run_fn" proprocessing_file = "modules/preprocessing.py" preprocessing_fn = "modules.preprocessing.preprocessing_fn" schema_file = "schema.pbtxt"import tfx tfx.__version__from tfx import v1 as tfx from tfx.orchestration.experimental.interactive.interactive_context import InteractiveContext from tfx.components import ImportExampleGen from tfx.components import StatisticsGen from tfx.components import SchemaGen from tfx.components import ExampleValidator from tfx.components import Transform from tfx.components import Trainer from tfx.components import Pusher from tfx.proto import example_gen_pb2context = InteractiveContext()input_config = example_gen_pb2.Input( splits=[ example_gen_pb2.Input.Split(name="train", pattern="train-*.tfrec"), example_gen_pb2.Input.Split(name="eval", pattern="val-*.tfrec"), ] ) example_gen = ImportExampleGen( input_base=local_data_path, input_config=input_config ) context.run(example_gen)statistics_gen = StatisticsGen(examples=example_gen.outputs['examples']) context.run(statistics_gen)context.show(statistics_gen.outputs['statistics'])%%writefile {schema_file} feature { name: "image" type: FLOAT presence { min_fraction: 1.0 } float_domain { min: 0 max: 255 } shape { dim { size: 256 } dim { size: 256 } dim { size: 3 } } } feature { name: "image_shape" type: INT presence { min_fraction: 1.0 } shape { dim { size: 3 } } } feature { name: "label" type: INT presence { min_fraction: 1.0 } int_domain { min: 0 max: 2 } shape { dim { size: 1 } } }schema_gen = tfx.components.ImportSchemaGen( schema_file=schema_file) context.run(schema_gen)example_validator = ExampleValidator( statistics=statistics_gen.outputs['statistics'], schema=schema_gen.outputs['schema'] ) context.run(example_validator)context.show(example_validator.outputs['anomalies'])%%writefile {proprocessing_file} import tensorflow as tf IMAGE_KEY = "image" LABEL_KEY = "label" MODEL_INPUT_IMAGE_KEY = "pixel_values" MODEL_INPUT_LABEL_KEY = "labels" INPUT_IMG_SIZE = 224 def preprocessing_fn(inputs): """tf.transform's callback function for preprocessing inputs. Args: inputs: map from feature keys to raw not-yet-transformed features. Returns: Map from string feature key to transformed feature operations. """ # print(inputs) outputs = {} inputs[IMAGE_KEY] = tf.image.resize( inputs[IMAGE_KEY], [INPUT_IMG_SIZE, INPUT_IMG_SIZE] ) inputs[IMAGE_KEY] = inputs[IMAGE_KEY] / 255.0 inputs[IMAGE_KEY] = tf.transpose(inputs[IMAGE_KEY], [0, 3, 1, 2]) outputs[MODEL_INPUT_IMAGE_KEY] = inputs[IMAGE_KEY] outputs[MODEL_INPUT_LABEL_KEY] = inputs[LABEL_KEY] return outputstransform = Transform( examples=example_gen.outputs['examples'], schema=schema_gen.outputs['schema'], preprocessing_fn=preprocessing_fn)context.run(transform)%%writefile {model_file} from typing import List, Dict, Tuple import absl import tensorflow as tf import tensorflow_transform as tft from transformers import ViTFeatureExtractor, TFViTForImageClassification from tfx.components.trainer.fn_args_utils import FnArgs from tfx_bsl.tfxio import dataset_options from tfx.components.trainer.fn_args_utils import DataAccessor feature_extractor = ViTFeatureExtractor() _TRAIN_LENGTH = 128 _EVAL_LENGTH = 128 _TRAIN_BATCH_SIZE = 8 _EVAL_BATCH_SIZE = 8 _EPOCHS = 1 _LABELS = ['angular_leaf_spot', 'bean_rust', 'healthy'] _CONCRETE_INPUT = "pixel_values" _MODEL_INPUT_LABEL_KEY = "labels" def INFO(text: str): absl.logging.info(text) def _normalize_img( img, mean=feature_extractor.image_mean, std=feature_extractor.image_std ): img = img / 255 mean = tf.constant(mean) std = tf.constant(std) return (img - mean) / std def _preprocess_serving(string_input): decoded_input = tf.io.decode_base64(string_input) decoded = tf.io.decode_jpeg(decoded_input, channels=3) resized = tf.image.resize(decoded, size=(224, 224)) normalized = _normalize_img(resized) normalized = tf.transpose( normalized, (2, 0, 1) ) # Since HF models are channel-first. return normalized @tf.function(input_signature=[tf.TensorSpec([None], tf.string)]) def _preprocess_fn(string_input): decoded_images = tf.map_fn( _preprocess_serving, string_input, dtype=tf.float32, back_prop=False ) return {_CONCRETE_INPUT: decoded_images} def _model_exporter(model: tf.keras.Model): m_call = tf.function(model.call).get_concrete_function( tf.TensorSpec( shape=[None, 3, 224, 224], dtype=tf.float32, name=_CONCRETE_INPUT ) ) @tf.function(input_signature=[tf.TensorSpec([None], tf.string)]) def serving_fn(string_input): labels = tf.constant(list(model.config.id2label.values()), dtype=tf.string) images = _preprocess_fn(string_input) predictions = m_call(**images) indices = tf.argmax(predictions.logits, axis=1) pred_source = tf.gather(params=labels, indices=indices) probs = tf.nn.softmax(predictions.logits, axis=1) pred_confidence = tf.reduce_max(probs, axis=1) return {"label": pred_source, "confidence": pred_confidence} return serving_fn def _input_fn( file_pattern: List[str], data_accessor: DataAccessor, tf_transform_output: tft.TFTransformOutput, is_train: bool = False, batch_size: int = 32, ) -> tf.data.Dataset: INFO(f"Reading data from: {file_pattern}") dataset = data_accessor.tf_dataset_factory( file_pattern, dataset_options.TensorFlowDatasetOptions( batch_size=batch_size, label_key=_MODEL_INPUT_LABEL_KEY ), tf_transform_output.transformed_metadata.schema, ) return dataset def _build_model(): id2label={str(i): c for i, c in enumerate(_LABELS)} label2id={c: str(i) for i, c in enumerate(_LABELS)} model = TFViTForImageClassification.from_pretrained( "google/vit-base-patch16-224-in21k", num_labels=len(_LABELS), label2id=label2id, id2label=id2label, ) model.layers[0].trainable=False loss = tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True) model.compile(optimizer='adam', loss=loss, metrics=["accuracy"]) return model def run_fn(fn_args: FnArgs): tf_transform_output = tft.TFTransformOutput(fn_args.transform_output) train_dataset = _input_fn( fn_args.train_files, fn_args.data_accessor, tf_transform_output, is_train=True, batch_size=_TRAIN_BATCH_SIZE, ) eval_dataset = _input_fn( fn_args.eval_files, fn_args.data_accessor, tf_transform_output, is_train=False, batch_size=_EVAL_BATCH_SIZE, ) model = _build_model() model.fit( train_dataset, steps_per_epoch=_TRAIN_LENGTH // _TRAIN_BATCH_SIZE, validation_data=eval_dataset, validation_steps=_EVAL_LENGTH // _EVAL_BATCH_SIZE, epochs=_EPOCHS, ) model.save( fn_args.serving_model_dir, save_format="tf", signatures=_model_exporter(model) )trainer = Trainer( run_fn=model_fn, transformed_examples=transform.outputs["transformed_examples"], transform_graph=transform.outputs["transform_graph"], schema=schema_gen.outputs["schema"], )context.run(trainer)
deep-diver/mlops-hf-tf-vision-models
notebooks/parse_tfrecord.ipynb
import tensorflow as tfGCS_PATH_FULL_RESOUTION = "gs://beans-fullres/tfrecords" GCS_PATH_LOW_RESOLUTION = "gs://beans-lowres/tfrecords"BATCH_SIZE = 4 AUTO = tf.data.AUTOTUNEdef parse_tfr(proto): feature_description = { "image": tf.io.VarLenFeature(tf.float32), "image_shape": tf.io.VarLenFeature(tf.int64), "label": tf.io.VarLenFeature(tf.int64), } rec = tf.io.parse_single_example(proto, feature_description) image_shape = tf.sparse.to_dense(rec["image_shape"]) image = tf.reshape(tf.sparse.to_dense(rec["image"]), image_shape) label = tf.sparse.to_dense(rec["label"]) return {"pixel_values": image, "label": label} def prepare_dataset(GCS_PATH=GCS_PATH_FULL_RESOUTION, split="train", batch_size=BATCH_SIZE): if split not in ["train", "val"]: raise ValueError( "Invalid split provided. Supports splits are: `train` and `val`." ) dataset = tf.data.TFRecordDataset( [filename for filename in tf.io.gfile.glob(f"{GCS_PATH}/{split}-*")], num_parallel_reads=AUTO, ).map(parse_tfr, num_parallel_calls=AUTO) if split == "train": dataset = dataset.shuffle(batch_size * 2) dataset = dataset.batch(batch_size) dataset = dataset.prefetch(AUTO) return datasettrain_dataset = prepare_dataset() val_dataset = prepare_dataset(split="val")for batch in train_dataset.take(1): print(batch["pixel_values"].shape, batch["label"].shape)for batch in val_dataset.take(1): print(batch["pixel_values"].shape, batch["label"].shape)train_dataset = prepare_dataset(GCS_PATH_LOW_RESOLUTION) val_dataset = prepare_dataset(GCS_PATH_LOW_RESOLUTION, split="val")for batch in train_dataset.take(1): print(batch["pixel_values"].shape, batch["label"].shape)for batch in val_dataset.take(1): print(batch["pixel_values"].shape, batch["label"].shape)
deep-diver/mlops-hf-tf-vision-models
advanced_part1/modules/ViT.py
import tensorflow as tf from transformers import TFViTForImageClassification from .common import LABELS from .common import PRETRAIN_CHECKPOINT from .utils import INFO def build_model(): id2label = {str(i): c for i, c in enumerate(LABELS)} label2id = {c: str(i) for i, c in enumerate(LABELS)} model = TFViTForImageClassification.from_pretrained( PRETRAIN_CHECKPOINT, num_labels=len(LABELS), label2id=label2id, id2label=id2label, ) model.layers[0].trainable = False loss = tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True) model.compile(optimizer="adam", loss=loss, metrics=["accuracy"]) INFO(model.summary()) return model
deep-diver/mlops-hf-tf-vision-models
advanced_part1/modules/common.py
IMAGE_TFREC_KEY = "image" IMAGE_SHAPE_TFREC_KEY = "image_shape" LABEL_TFREC_KEY = "label" MODEL_INPUT_IMAGE_KEY = "pixel_values" MODEL_INPUT_LABEL_KEY = "labels" IMAGE_MODEL_KEY = "pixel_values" LABEL_MODEL_KEY = "labels" CONCRETE_INPUT = "pixel_values" PRETRAIN_CHECKPOINT = "google/vit-base-patch16-224-in21k" LABELS = ["angular_leaf_spot", "bean_rust", "healthy"]
deep-diver/mlops-hf-tf-vision-models
advanced_part1/modules/hyperparams.py
EPOCHS = 1 BATCH_SIZE = 32 TRAIN_BATCH_SIZE = 32 EVAL_BATCH_SIZE = 32 TRAIN_LENGTH = 1034 EVAL_LENGTH = 128 INPUT_IMG_SIZE = 224
deep-diver/mlops-hf-tf-vision-models
advanced_part1/modules/preprocessing.py
import tensorflow as tf from .common import IMAGE_TFREC_KEY, LABEL_TFREC_KEY from .common import IMAGE_MODEL_KEY, LABEL_MODEL_KEY from .hyperparams import INPUT_IMG_SIZE def preprocessing_fn(inputs): """tf.transform's callback function for preprocessing inputs. Args: inputs: map from feature keys to raw not-yet-transformed features. Returns: Map from string feature key to transformed feature operations. """ # print(inputs) outputs = {} inputs[IMAGE_TFREC_KEY] = tf.image.resize( inputs[IMAGE_TFREC_KEY], [INPUT_IMG_SIZE, INPUT_IMG_SIZE] ) inputs[IMAGE_TFREC_KEY] = inputs[IMAGE_TFREC_KEY] / 255.0 inputs[IMAGE_TFREC_KEY] = tf.transpose(inputs[IMAGE_TFREC_KEY], [0, 3, 1, 2]) outputs[IMAGE_MODEL_KEY] = inputs[IMAGE_TFREC_KEY] outputs[LABEL_MODEL_KEY] = inputs[LABEL_TFREC_KEY] return outputs
deep-diver/mlops-hf-tf-vision-models
advanced_part1/modules/signatures.py
from typing import Dict import tensorflow as tf import tensorflow_transform as tft from transformers import ViTFeatureExtractor from .common import PRETRAIN_CHECKPOINT from .common import CONCRETE_INPUT from .common import LABEL_MODEL_KEY feature_extractor = ViTFeatureExtractor.from_pretrained(PRETRAIN_CHECKPOINT) def _normalize_img( img, mean=feature_extractor.image_mean, std=feature_extractor.image_std ): img = img / 255 mean = tf.constant(mean) std = tf.constant(std) return (img - mean) / std def _preprocess_serving(string_input): decoded_input = tf.io.decode_base64(string_input) decoded = tf.io.decode_jpeg(decoded_input, channels=3) resized = tf.image.resize(decoded, size=(224, 224)) normalized = _normalize_img(resized) normalized = tf.transpose( normalized, (2, 0, 1) ) # Since HF models are channel-first. return normalized @tf.function(input_signature=[tf.TensorSpec([None], tf.string)]) def _preprocess_fn(string_input): decoded_images = tf.map_fn( _preprocess_serving, string_input, dtype=tf.float32, back_prop=False ) return {CONCRETE_INPUT: decoded_images} def model_exporter(model: tf.keras.Model): m_call = tf.function(model.call).get_concrete_function( tf.TensorSpec(shape=[None, 3, 224, 224], dtype=tf.float32, name=CONCRETE_INPUT) ) @tf.function(input_signature=[tf.TensorSpec([None], tf.string)]) def serving_fn(string_input): labels = tf.constant(list(model.config.id2label.values()), dtype=tf.string) images = _preprocess_fn(string_input) predictions = m_call(**images) indices = tf.argmax(predictions.logits, axis=1) pred_source = tf.gather(params=labels, indices=indices) probs = tf.nn.softmax(predictions.logits, axis=1) pred_confidence = tf.reduce_max(probs, axis=1) return {"label": pred_source, "confidence": pred_confidence} return serving_fn def transform_features_signature( model: tf.keras.Model, tf_transform_output: tft.TFTransformOutput ): """ transform_features_signature simply returns a function that transforms any data of the type of tf.Example which is denoted as the type of sta ndard_artifacts.Examples in TFX. The purpose of this function is to ap ply Transform Graph obtained from Transform component to the data prod uced by ImportExampleGen. This function will be used in the Evaluator component, so the raw evaluation inputs from ImportExampleGen can be a pporiately transformed that the model could understand. """ # basically, what Transform component emits is a SavedModel that knows # how to transform data. transform_features_layer() simply returns the # layer from the Transform. model.tft_layer = tf_transform_output.transform_features_layer() @tf.function( input_signature=[tf.TensorSpec(shape=[None], dtype=tf.string, name="examples")] ) def serve_tf_examples_fn(serialized_tf_examples): """ raw_feature_spec returns a set of feature maps(dict) for the input TFRecords based on the knowledge that Transform component has lear ned(learn doesn't mean training here). By using this information, the raw data from ImportExampleGen could be parsed with tf.io.parse _example utility function. Then, it is passed to the model.tft_layer, so the final output we get is the transformed data of the raw input. """ feature_spec = tf_transform_output.raw_feature_spec() parsed_features = tf.io.parse_example(serialized_tf_examples, feature_spec) transformed_features = model.tft_layer(parsed_features) return transformed_features return serve_tf_examples_fn def tf_examples_serving_signature(model, tf_transform_output): """ tf_examples_serving_signature simply returns a function that performs data transformation(preprocessing) and model prediction in a sequential manner. How data transformation is done is idential to the process of transform_features_signature function. """ model.tft_layer = tf_transform_output.transform_features_layer() @tf.function( input_signature=[tf.TensorSpec(shape=[None], dtype=tf.string, name="examples")] ) def serve_tf_examples_fn( serialized_tf_example: tf.Tensor, ) -> Dict[str, tf.Tensor]: raw_feature_spec = tf_transform_output.raw_feature_spec() raw_features = tf.io.parse_example(serialized_tf_example, raw_feature_spec) transformed_features = model.tft_layer(raw_features) logits = model(transformed_features).logits return {LABEL_MODEL_KEY: logits} return serve_tf_examples_fn
deep-diver/mlops-hf-tf-vision-models
advanced_part1/modules/train.py
import tensorflow_transform as tft from tfx.components.trainer.fn_args_utils import FnArgs from .train_data import input_fn from .ViT import build_model from .signatures import ( model_exporter, transform_features_signature, tf_examples_serving_signature, ) from .hyperparams import TRAIN_BATCH_SIZE, EVAL_BATCH_SIZE from .hyperparams import TRAIN_LENGTH, EVAL_LENGTH from .hyperparams import EPOCHS def run_fn(fn_args: FnArgs): tf_transform_output = tft.TFTransformOutput(fn_args.transform_output) train_dataset = input_fn( fn_args.train_files, fn_args.data_accessor, tf_transform_output, is_train=True, batch_size=TRAIN_BATCH_SIZE, ) eval_dataset = input_fn( fn_args.eval_files, fn_args.data_accessor, tf_transform_output, is_train=False, batch_size=EVAL_BATCH_SIZE, ) model = build_model() model.fit( train_dataset, steps_per_epoch=TRAIN_LENGTH // TRAIN_BATCH_SIZE, validation_data=eval_dataset, validation_steps=EVAL_LENGTH // TRAIN_BATCH_SIZE, epochs=EPOCHS, ) model.save( fn_args.serving_model_dir, save_format="tf", signatures={ "serving_default": model_exporter(model), "transform_features": transform_features_signature( model, tf_transform_output ), "from_examples": tf_examples_serving_signature(model, tf_transform_output), }, )
deep-diver/mlops-hf-tf-vision-models
advanced_part1/modules/train_data.py
from typing import List import tensorflow as tf import tensorflow_transform as tft from tfx_bsl.tfxio import dataset_options from tfx.components.trainer.fn_args_utils import DataAccessor from .utils import INFO from .common import LABEL_MODEL_KEY from .hyperparams import BATCH_SIZE def input_fn( file_pattern: List[str], data_accessor: DataAccessor, tf_transform_output: tft.TFTransformOutput, is_train: bool = False, batch_size: int = BATCH_SIZE, ) -> tf.data.Dataset: INFO(f"Reading data from: {file_pattern}") dataset = data_accessor.tf_dataset_factory( file_pattern, dataset_options.TensorFlowDatasetOptions( batch_size=batch_size, label_key=LABEL_MODEL_KEY ), tf_transform_output.transformed_metadata.schema, ) return dataset
deep-diver/mlops-hf-tf-vision-models
advanced_part1/modules/utils.py
import absl def INFO(text: str): absl.logging.info(text)
deep-diver/mlops-hf-tf-vision-models
advanced_part1/pipeline/configs.py
import os import tensorflow_model_analysis as tfma import tfx.extensions.google_cloud_ai_platform.constants as vertex_const import tfx.extensions.google_cloud_ai_platform.trainer.executor as vertex_training_const PIPELINE_NAME = "vit-e2e-pipeline-advanced-part1" try: import google.auth # pylint: disable=g-import-not-at-top # pytype: disable=import-error try: _, GOOGLE_CLOUD_PROJECT = google.auth.default() except google.auth.exceptions.DefaultCredentialsError: GOOGLE_CLOUD_PROJECT = "gcp-ml-172005" except ImportError: GOOGLE_CLOUD_PROJECT = "gcp-ml-172005" GOOGLE_CLOUD_REGION = "us-central1" GCS_BUCKET_NAME = GOOGLE_CLOUD_PROJECT + "-complete-mlops" PIPELINE_IMAGE = f"gcr.io/{GOOGLE_CLOUD_PROJECT}/{PIPELINE_NAME}" OUTPUT_DIR = os.path.join("gs://", GCS_BUCKET_NAME) PIPELINE_ROOT = os.path.join(OUTPUT_DIR, "tfx_pipeline_output", PIPELINE_NAME) DATA_PATH = "gs://beans-lowres/tfrecords/" SCHEMA_PATH = "pipeline/schema.pbtxt" TRAINING_FN = "modules.train.run_fn" PREPROCESSING_FN = "modules.preprocessing.preprocessing_fn" EXAMPLE_GEN_BEAM_ARGS = None TRANSFORM_BEAM_ARGS = None EVAL_CONFIGS = tfma.EvalConfig( model_specs=[ tfma.ModelSpec( signature_name="from_examples", preprocessing_function_names=["transform_features"], label_key="labels", prediction_key="labels", ) ], slicing_specs=[tfma.SlicingSpec()], metrics_specs=[ tfma.MetricsSpec( metrics=[ tfma.MetricConfig( class_name="SparseCategoricalAccuracy", threshold=tfma.MetricThreshold( value_threshold=tfma.GenericValueThreshold( lower_bound={"value": 0.55} ), # Change threshold will be ignored if there is no # baseline model resolved from MLMD (first run). change_threshold=tfma.GenericChangeThreshold( direction=tfma.MetricDirection.HIGHER_IS_BETTER, absolute={"value": -1e-3}, ), ), ) ] ) ], ) GCP_AI_PLATFORM_TRAINING_ARGS = { vertex_const.ENABLE_VERTEX_KEY: True, vertex_const.VERTEX_REGION_KEY: GOOGLE_CLOUD_REGION, vertex_training_const.TRAINING_ARGS_KEY: { "project": GOOGLE_CLOUD_PROJECT, "worker_pool_specs": [ { "machine_spec": { "machine_type": "n1-standard-4", "accelerator_type": "NVIDIA_TESLA_K80", "accelerator_count": 1, }, "replica_count": 1, "container_spec": { "image_uri": PIPELINE_IMAGE, }, } ], }, "use_gpu": True, } fullres_data = os.environ.get("FULL_RES_DATA", "false") if fullres_data.lower() == "true": DATA_PATH = "gs://beans-fullres/tfrecords/" DATAFLOW_SERVICE_ACCOUNT = "csp-gde-dataflow@gcp-ml-172005.iam.gserviceaccount.com" DATAFLOW_MACHINE_TYPE = "n1-standard-4" DATAFLOW_MAX_WORKERS = 4 DATAFLOW_DISK_SIZE_GB = 100 EXAMPLE_GEN_BEAM_ARGS = [ "--runner=DataflowRunner", "--project=" + GOOGLE_CLOUD_PROJECT, "--region=" + GOOGLE_CLOUD_REGION, "--service_account_email=" + DATAFLOW_SERVICE_ACCOUNT, "--machine_type=" + DATAFLOW_MACHINE_TYPE, "--experiments=use_runner_v2", "--max_num_workers=" + str(DATAFLOW_MAX_WORKERS), "--disk_size_gb=" + str(DATAFLOW_DISK_SIZE_GB), ] TRANSFORM_BEAM_ARGS = [ "--runner=DataflowRunner", "--project=" + GOOGLE_CLOUD_PROJECT, "--region=" + GOOGLE_CLOUD_REGION, "--service_account_email=" + DATAFLOW_SERVICE_ACCOUNT, "--machine_type=" + DATAFLOW_MACHINE_TYPE, "--experiments=use_runner_v2", "--max_num_workers=" + str(DATAFLOW_MAX_WORKERS), "--disk_size_gb=" + str(DATAFLOW_DISK_SIZE_GB), "--worker_harness_container_image=" + PIPELINE_IMAGE, ] GCP_AI_PLATFORM_TRAINING_ARGS[vertex_training_const.TRAINING_ARGS_KEY][ "worker_pool_specs" ] = [ { "machine_spec": { "machine_type": "n1-standard-8", "accelerator_type": "NVIDIA_TESLA_V100", "accelerator_count": 1, }, "replica_count": 1, "container_spec": { "image_uri": PIPELINE_IMAGE, }, } ] GCP_AI_PLATFORM_SERVING_ARGS = { vertex_const.ENABLE_VERTEX_KEY: True, vertex_const.VERTEX_REGION_KEY: GOOGLE_CLOUD_REGION, vertex_const.VERTEX_CONTAINER_IMAGE_URI_KEY: "us-docker.pkg.dev/vertex-ai/prediction/tf2-cpu.2-8:latest", vertex_const.SERVING_ARGS_KEY: { "project_id": GOOGLE_CLOUD_PROJECT, "deployed_model_display_name": PIPELINE_NAME.replace("-", "_"), "endpoint_name": "prediction-" + PIPELINE_NAME.replace("-", "_"), "traffic_split": {"0": 100}, "machine_type": "n1-standard-4", "min_replica_count": 1, "max_replica_count": 1, }, }
deep-diver/mlops-hf-tf-vision-models
advanced_part1/pipeline/kubeflow_pipeline.py
from typing import Any, Dict, List, Optional, Text import tensorflow_model_analysis as tfma from tfx import v1 as tfx from ml_metadata.proto import metadata_store_pb2 from tfx.proto import example_gen_pb2 from tfx.components import ImportExampleGen from tfx.components import StatisticsGen from tfx.components import ExampleValidator from tfx.components import Transform from tfx.components import Evaluator from tfx.extensions.google_cloud_ai_platform.trainer.component import ( Trainer as VertexTrainer, ) from tfx.extensions.google_cloud_ai_platform.pusher.component import ( Pusher as VertexPusher, ) from tfx.orchestration import pipeline from tfx.proto import example_gen_pb2 from tfx.types import Channel from tfx.types.standard_artifacts import Model from tfx.types.standard_artifacts import ModelBlessing from tfx.dsl.components.common import resolver from tfx.dsl.experimental.latest_blessed_model_resolver import ( LatestBlessedModelResolver, ) def create_pipeline( pipeline_name: Text, pipeline_root: Text, data_path: Text, schema_path: Text, modules: Dict[Text, Text], eval_configs: tfma.EvalConfig, metadata_connection_config: Optional[metadata_store_pb2.ConnectionConfig] = None, ai_platform_training_args: Optional[Dict[Text, Text]] = None, ai_platform_serving_args: Optional[Dict[Text, Any]] = None, example_gen_beam_args: Optional[List] = None, transform_beam_args: Optional[List] = None, ) -> tfx.dsl.Pipeline: components = [] input_config = example_gen_pb2.Input( splits=[ example_gen_pb2.Input.Split(name="train", pattern="train-*.tfrec"), example_gen_pb2.Input.Split(name="eval", pattern="val-*.tfrec"), ] ) example_gen = ImportExampleGen(input_base=data_path, input_config=input_config) if example_gen_beam_args is not None: example_gen.with_beam_pipeline_args(example_gen_beam_args) components.append(example_gen) statistics_gen = StatisticsGen(examples=example_gen.outputs["examples"]) components.append(statistics_gen) schema_gen = tfx.components.ImportSchemaGen(schema_file=schema_path) components.append(schema_gen) example_validator = ExampleValidator( statistics=statistics_gen.outputs["statistics"], schema=schema_gen.outputs["schema"], ) components.append(example_validator) transform_args = { "examples": example_gen.outputs["examples"], "schema": schema_gen.outputs["schema"], "preprocessing_fn": modules["preprocessing_fn"], } transform = Transform(**transform_args) if transform_beam_args is not None: transform.with_beam_pipeline_args(transform_beam_args) components.append(transform) trainer_args = { "run_fn": modules["training_fn"], "transformed_examples": transform.outputs["transformed_examples"], "transform_graph": transform.outputs["transform_graph"], "schema": schema_gen.outputs["schema"], "custom_config": ai_platform_training_args, } trainer = VertexTrainer(**trainer_args) components.append(trainer) model_resolver = resolver.Resolver( strategy_class=LatestBlessedModelResolver, model=Channel(type=Model), model_blessing=Channel(type=ModelBlessing), ).with_id("latest_blessed_model_resolver") components.append(model_resolver) evaluator = Evaluator( examples=example_gen.outputs["examples"], model=trainer.outputs["model"], baseline_model=model_resolver.outputs["model"], eval_config=eval_configs, ) components.append(evaluator) pusher_args = { "model": trainer.outputs["model"], "model_blessing": evaluator.outputs["blessing"], "custom_config": ai_platform_serving_args, } pusher = VertexPusher(**pusher_args) # pylint: disable=unused-variable components.append(pusher) return pipeline.Pipeline( pipeline_name=pipeline_name, pipeline_root=pipeline_root, components=components, enable_cache=True, metadata_connection_config=metadata_connection_config, )
deep-diver/mlops-hf-tf-vision-models
advanced_part1/pipeline/local_pipeline.py
from typing import Dict, Optional, Text import tensorflow_model_analysis as tfma from tfx import v1 as tfx from ml_metadata.proto import metadata_store_pb2 from tfx.proto import example_gen_pb2 from tfx.components import ImportExampleGen from tfx.components import StatisticsGen from tfx.components import ExampleValidator from tfx.components import Transform from tfx.components import Trainer from tfx.components import Evaluator from tfx.components import Pusher from tfx.orchestration import pipeline from tfx.proto import example_gen_pb2 from tfx.types import Channel from tfx.types.standard_artifacts import Model from tfx.types.standard_artifacts import ModelBlessing from tfx.dsl.components.common import resolver from tfx.dsl.experimental.latest_blessed_model_resolver import ( LatestBlessedModelResolver, ) def create_pipeline( pipeline_name: Text, pipeline_root: Text, data_path: Text, schema_path: Text, modules: Dict[Text, Text], eval_configs: tfma.EvalConfig, serving_model_dir: Text, metadata_connection_config: Optional[metadata_store_pb2.ConnectionConfig] = None, ) -> tfx.dsl.Pipeline: components = [] input_config = example_gen_pb2.Input( splits=[ example_gen_pb2.Input.Split(name="train", pattern="train-00-*.tfrec"), example_gen_pb2.Input.Split(name="eval", pattern="val-00-*.tfrec"), ] ) example_gen = ImportExampleGen(input_base=data_path, input_config=input_config) components.append(example_gen) statistics_gen = StatisticsGen(examples=example_gen.outputs["examples"]) components.append(statistics_gen) schema_gen = tfx.components.ImportSchemaGen(schema_file=schema_path) components.append(schema_gen) example_validator = ExampleValidator( statistics=statistics_gen.outputs["statistics"], schema=schema_gen.outputs["schema"], ) components.append(example_validator) transform_args = { "examples": example_gen.outputs["examples"], "schema": schema_gen.outputs["schema"], "preprocessing_fn": modules["preprocessing_fn"], } transform = Transform(**transform_args) components.append(transform) trainer_args = { "run_fn": modules["training_fn"], "transformed_examples": transform.outputs["transformed_examples"], "transform_graph": transform.outputs["transform_graph"], "schema": schema_gen.outputs["schema"], } trainer = Trainer(**trainer_args) components.append(trainer) model_resolver = resolver.Resolver( strategy_class=LatestBlessedModelResolver, model=Channel(type=Model), model_blessing=Channel(type=ModelBlessing), ).with_id("latest_blessed_model_resolver") components.append(model_resolver) evaluator = Evaluator( examples=example_gen.outputs["examples"], model=trainer.outputs["model"], baseline_model=model_resolver.outputs["model"], eval_config=eval_configs, ) components.append(evaluator) pusher_args = { "model": trainer.outputs["model"], "model_blessing": evaluator.outputs["blessing"], "push_destination": tfx.proto.PushDestination( filesystem=tfx.proto.PushDestination.Filesystem( base_directory=serving_model_dir ) ), } pusher = Pusher(**pusher_args) # pylint: disable=unused-variable components.append(pusher) return pipeline.Pipeline( pipeline_name=pipeline_name, pipeline_root=pipeline_root, components=components, enable_cache=False, metadata_connection_config=metadata_connection_config, )
deep-diver/mlops-hf-tf-vision-models
advanced_part2/modules/ViT.py
import tensorflow as tf import keras_tuner from transformers import TFViTForImageClassification from .common import LABELS from .common import PRETRAIN_CHECKPOINT from .utils import INFO def build_model(hparams: keras_tuner.HyperParameters): id2label = {str(i): c for i, c in enumerate(LABELS)} label2id = {c: str(i) for i, c in enumerate(LABELS)} model = TFViTForImageClassification.from_pretrained( PRETRAIN_CHECKPOINT, num_labels=len(LABELS), label2id=label2id, id2label=id2label, ) model.layers[0].trainable = False loss = tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True) optimizer = tf.keras.optimizers.Adam(learning_rate=hparams.get("learning_rate")) model.compile(optimizer=optimizer, loss=loss, metrics=["accuracy"]) INFO(model.summary()) return model
deep-diver/mlops-hf-tf-vision-models
advanced_part2/modules/common.py
IMAGE_TFREC_KEY = "image" IMAGE_SHAPE_TFREC_KEY = "image_shape" LABEL_TFREC_KEY = "label" MODEL_INPUT_IMAGE_KEY = "pixel_values" MODEL_INPUT_LABEL_KEY = "labels" IMAGE_MODEL_KEY = "pixel_values" LABEL_MODEL_KEY = "labels" CONCRETE_INPUT = "pixel_values" PRETRAIN_CHECKPOINT = "google/vit-base-patch16-224-in21k" LABELS = ["angular_leaf_spot", "bean_rust", "healthy"]
deep-diver/mlops-hf-tf-vision-models
advanced_part2/modules/hyperparams.py
import keras_tuner EPOCHS = 10 BATCH_SIZE = 32 TRAIN_BATCH_SIZE = 32 EVAL_BATCH_SIZE = 32 TRAIN_LENGTH = 1034 EVAL_LENGTH = 128 INPUT_IMG_SIZE = 224 def get_hyperparameters(hyperparameters) -> keras_tuner.HyperParameters: hp_set = keras_tuner.HyperParameters() for hp in hyperparameters: hp_set.Choice( hp, hyperparameters[hp]["values"], default=hyperparameters[hp]["default"] ) return hp_set
deep-diver/mlops-hf-tf-vision-models
advanced_part2/modules/preprocessing.py
import tensorflow as tf from .common import IMAGE_TFREC_KEY, LABEL_TFREC_KEY from .common import IMAGE_MODEL_KEY, LABEL_MODEL_KEY from .hyperparams import INPUT_IMG_SIZE def preprocessing_fn(inputs): """tf.transform's callback function for preprocessing inputs. Args: inputs: map from feature keys to raw not-yet-transformed features. Returns: Map from string feature key to transformed feature operations. """ # print(inputs) outputs = {} inputs[IMAGE_TFREC_KEY] = tf.image.resize( inputs[IMAGE_TFREC_KEY], [INPUT_IMG_SIZE, INPUT_IMG_SIZE] ) inputs[IMAGE_TFREC_KEY] = inputs[IMAGE_TFREC_KEY] / 255.0 inputs[IMAGE_TFREC_KEY] = tf.transpose(inputs[IMAGE_TFREC_KEY], [0, 3, 1, 2]) outputs[IMAGE_MODEL_KEY] = inputs[IMAGE_TFREC_KEY] outputs[LABEL_MODEL_KEY] = inputs[LABEL_TFREC_KEY] return outputs
deep-diver/mlops-hf-tf-vision-models
advanced_part2/modules/signatures.py
from typing import Dict import tensorflow as tf import tensorflow_transform as tft from transformers import ViTFeatureExtractor from .common import PRETRAIN_CHECKPOINT from .common import CONCRETE_INPUT from .common import LABEL_MODEL_KEY feature_extractor = ViTFeatureExtractor.from_pretrained(PRETRAIN_CHECKPOINT) def _normalize_img( img, mean=feature_extractor.image_mean, std=feature_extractor.image_std ): img = img / 255 mean = tf.constant(mean) std = tf.constant(std) return (img - mean) / std def _preprocess_serving(string_input): decoded_input = tf.io.decode_base64(string_input) decoded = tf.io.decode_jpeg(decoded_input, channels=3) resized = tf.image.resize(decoded, size=(224, 224)) normalized = _normalize_img(resized) normalized = tf.transpose( normalized, (2, 0, 1) ) # Since HF models are channel-first. return normalized @tf.function(input_signature=[tf.TensorSpec([None], tf.string)]) def _preprocess_fn(string_input): decoded_images = tf.map_fn( _preprocess_serving, string_input, dtype=tf.float32, back_prop=False ) return {CONCRETE_INPUT: decoded_images} def model_exporter(model: tf.keras.Model): m_call = tf.function(model.call).get_concrete_function( tf.TensorSpec(shape=[None, 3, 224, 224], dtype=tf.float32, name=CONCRETE_INPUT) ) @tf.function(input_signature=[tf.TensorSpec([None], tf.string)]) def serving_fn(string_input): labels = tf.constant(list(model.config.id2label.values()), dtype=tf.string) images = _preprocess_fn(string_input) predictions = m_call(**images) indices = tf.argmax(predictions.logits, axis=1) pred_source = tf.gather(params=labels, indices=indices) probs = tf.nn.softmax(predictions.logits, axis=1) pred_confidence = tf.reduce_max(probs, axis=1) return {"label": pred_source, "confidence": pred_confidence} return serving_fn def transform_features_signature( model: tf.keras.Model, tf_transform_output: tft.TFTransformOutput ): """ transform_features_signature simply returns a function that transforms any data of the type of tf.Example which is denoted as the type of sta ndard_artifacts.Examples in TFX. The purpose of this function is to ap ply Transform Graph obtained from Transform component to the data prod uced by ImportExampleGen. This function will be used in the Evaluator component, so the raw evaluation inputs from ImportExampleGen can be a pporiately transformed that the model could understand. """ # basically, what Transform component emits is a SavedModel that knows # how to transform data. transform_features_layer() simply returns the # layer from the Transform. model.tft_layer = tf_transform_output.transform_features_layer() @tf.function( input_signature=[tf.TensorSpec(shape=[None], dtype=tf.string, name="examples")] ) def serve_tf_examples_fn(serialized_tf_examples): """ raw_feature_spec returns a set of feature maps(dict) for the input TFRecords based on the knowledge that Transform component has lear ned(learn doesn't mean training here). By using this information, the raw data from ImportExampleGen could be parsed with tf.io.parse _example utility function. Then, it is passed to the model.tft_layer, so the final output we get is the transformed data of the raw input. """ feature_spec = tf_transform_output.raw_feature_spec() parsed_features = tf.io.parse_example(serialized_tf_examples, feature_spec) transformed_features = model.tft_layer(parsed_features) return transformed_features return serve_tf_examples_fn def tf_examples_serving_signature(model, tf_transform_output): """ tf_examples_serving_signature simply returns a function that performs data transformation(preprocessing) and model prediction in a sequential manner. How data transformation is done is idential to the process of transform_features_signature function. """ model.tft_layer = tf_transform_output.transform_features_layer() @tf.function( input_signature=[tf.TensorSpec(shape=[None], dtype=tf.string, name="examples")] ) def serve_tf_examples_fn( serialized_tf_example: tf.Tensor, ) -> Dict[str, tf.Tensor]: raw_feature_spec = tf_transform_output.raw_feature_spec() raw_features = tf.io.parse_example(serialized_tf_example, raw_feature_spec) transformed_features = model.tft_layer(raw_features) logits = model(transformed_features).logits return {LABEL_MODEL_KEY: logits} return serve_tf_examples_fn
deep-diver/mlops-hf-tf-vision-models
advanced_part2/modules/train.py
import keras_tuner import tensorflow_transform as tft from tfx.components.trainer.fn_args_utils import FnArgs from .train_data import input_fn from .ViT import build_model from .signatures import ( model_exporter, transform_features_signature, tf_examples_serving_signature, ) from .hyperparams import TRAIN_BATCH_SIZE, EVAL_BATCH_SIZE from .hyperparams import TRAIN_LENGTH, EVAL_LENGTH from .hyperparams import EPOCHS from .utils import INFO def run_fn(fn_args: FnArgs): custom_config = fn_args.custom_config epochs = EPOCHS if custom_config is not None: if "is_local" in custom_config: epochs = 1 tf_transform_output = tft.TFTransformOutput(fn_args.transform_output) train_dataset = input_fn( fn_args.train_files, fn_args.data_accessor, tf_transform_output, is_train=True, batch_size=TRAIN_BATCH_SIZE, ) eval_dataset = input_fn( fn_args.eval_files, fn_args.data_accessor, tf_transform_output, is_train=False, batch_size=EVAL_BATCH_SIZE, ) hparams = keras_tuner.HyperParameters.from_config(fn_args.hyperparameters) INFO(f"HyperParameters for training: {hparams.get_config()}") model = build_model(hparams) model.fit( train_dataset, steps_per_epoch=TRAIN_LENGTH // TRAIN_BATCH_SIZE, validation_data=eval_dataset, validation_steps=EVAL_LENGTH // TRAIN_BATCH_SIZE, epochs=epochs, ) model.save( fn_args.serving_model_dir, save_format="tf", signatures={ "serving_default": model_exporter(model), "transform_features": transform_features_signature( model, tf_transform_output ), "from_examples": tf_examples_serving_signature(model, tf_transform_output), }, )
deep-diver/mlops-hf-tf-vision-models
advanced_part2/modules/train_data.py
from typing import List import tensorflow as tf import tensorflow_transform as tft from tfx_bsl.tfxio import dataset_options from tfx.components.trainer.fn_args_utils import DataAccessor from .utils import INFO from .common import LABEL_MODEL_KEY from .hyperparams import BATCH_SIZE def input_fn( file_pattern: List[str], data_accessor: DataAccessor, tf_transform_output: tft.TFTransformOutput, is_train: bool = False, batch_size: int = BATCH_SIZE, ) -> tf.data.Dataset: INFO(f"Reading data from: {file_pattern}") dataset = data_accessor.tf_dataset_factory( file_pattern, dataset_options.TensorFlowDatasetOptions( batch_size=batch_size, label_key=LABEL_MODEL_KEY ), tf_transform_output.transformed_metadata.schema, ) return dataset
deep-diver/mlops-hf-tf-vision-models
advanced_part2/modules/tuning.py
import keras_tuner import tensorflow_transform as tft from tfx.components.trainer.fn_args_utils import FnArgs from tfx.v1.components import TunerFnResult from .train_data import input_fn from .ViT import build_model from .hyperparams import TRAIN_BATCH_SIZE, EVAL_BATCH_SIZE from .hyperparams import TRAIN_LENGTH, EVAL_LENGTH from .hyperparams import get_hyperparameters def tuner_fn(fn_args: FnArgs) -> TunerFnResult: hyperparameters = fn_args.custom_config["hyperparameters"] tuner = keras_tuner.RandomSearch( build_model, max_trials=6, hyperparameters=get_hyperparameters(hyperparameters), allow_new_entries=False, objective=keras_tuner.Objective("val_accuracy", "max"), directory=fn_args.working_dir, project_name="ViT MLOps Advanced Part2", ) tf_transform_output = tft.TFTransformOutput(fn_args.transform_graph_path) train_dataset = input_fn( fn_args.train_files, fn_args.data_accessor, tf_transform_output, is_train=True, batch_size=TRAIN_BATCH_SIZE, ) eval_dataset = input_fn( fn_args.eval_files, fn_args.data_accessor, tf_transform_output, is_train=False, batch_size=EVAL_BATCH_SIZE, ) return TunerFnResult( tuner=tuner, fit_kwargs={ "x": train_dataset, "validation_data": eval_dataset, "steps_per_epoch": TRAIN_LENGTH // TRAIN_BATCH_SIZE, "validation_steps": EVAL_LENGTH // EVAL_BATCH_SIZE, }, )
deep-diver/mlops-hf-tf-vision-models
advanced_part2/modules/utils.py
import absl def INFO(text: str): absl.logging.info(text)
deep-diver/mlops-hf-tf-vision-models
advanced_part2/pipeline/configs.py
import os import tensorflow_model_analysis as tfma import tfx.extensions.google_cloud_ai_platform.constants as vertex_const import tfx.extensions.google_cloud_ai_platform.trainer.executor as vertex_training_const import tfx.extensions.google_cloud_ai_platform.tuner.executor as vertex_tuner_const PIPELINE_NAME = "vit-e2e-pipeline-advanced-part2" try: import google.auth # pylint: disable=g-import-not-at-top # pytype: disable=import-error try: _, GOOGLE_CLOUD_PROJECT = google.auth.default() except google.auth.exceptions.DefaultCredentialsError: GOOGLE_CLOUD_PROJECT = "gcp-ml-172005" except ImportError: GOOGLE_CLOUD_PROJECT = "gcp-ml-172005" GOOGLE_CLOUD_REGION = "us-central1" GCS_BUCKET_NAME = GOOGLE_CLOUD_PROJECT + "-complete-mlops" PIPELINE_IMAGE = f"gcr.io/{GOOGLE_CLOUD_PROJECT}/{PIPELINE_NAME}" OUTPUT_DIR = os.path.join("gs://", GCS_BUCKET_NAME) PIPELINE_ROOT = os.path.join(OUTPUT_DIR, "tfx_pipeline_output", PIPELINE_NAME) DATA_PATH = "gs://beans-lowres/tfrecords/" SCHEMA_PATH = "pipeline/schema.pbtxt" TRAINING_FN = "modules.train.run_fn" TUNER_FN = "modules.tuning.tuner_fn" PREPROCESSING_FN = "modules.preprocessing.preprocessing_fn" EXAMPLE_GEN_BEAM_ARGS = None TRANSFORM_BEAM_ARGS = None TRAIN_BATCH_SIZE = 32 EVAL_BATCH_SIZE = 32 TRAIN_LENGTH = 1034 EVAL_LENGTH = 128 HYPER_PARAMETERS = { "learning_rate": {"values": [1e-3, 1e-2, 1e-1], "default": 1e-3}, } EVAL_CONFIGS = tfma.EvalConfig( model_specs=[ tfma.ModelSpec( signature_name="from_examples", preprocessing_function_names=["transform_features"], label_key="labels", prediction_key="labels", ) ], slicing_specs=[tfma.SlicingSpec()], metrics_specs=[ tfma.MetricsSpec( metrics=[ tfma.MetricConfig( class_name="SparseCategoricalAccuracy", threshold=tfma.MetricThreshold( value_threshold=tfma.GenericValueThreshold( lower_bound={"value": 0.55} ), # Change threshold will be ignored if there is no # baseline model resolved from MLMD (first run). change_threshold=tfma.GenericChangeThreshold( direction=tfma.MetricDirection.HIGHER_IS_BETTER, absolute={"value": -1e-3}, ), ), ) ] ) ], ) GCP_AI_PLATFORM_TRAINING_ARGS = { vertex_const.ENABLE_VERTEX_KEY: True, vertex_const.VERTEX_REGION_KEY: GOOGLE_CLOUD_REGION, vertex_training_const.TRAINING_ARGS_KEY: { "project": GOOGLE_CLOUD_PROJECT, "worker_pool_specs": [ { "machine_spec": { "machine_type": "n1-standard-4", "accelerator_type": "NVIDIA_TESLA_K80", "accelerator_count": 1, }, "replica_count": 1, "container_spec": { "image_uri": PIPELINE_IMAGE, }, } ], }, "use_gpu": True, } fullres_data = os.environ.get("FULL_RES_DATA", "false") if fullres_data.lower() == "true": DATA_PATH = "gs://beans-fullres/tfrecords/" DATAFLOW_SERVICE_ACCOUNT = "csp-gde-dataflow@gcp-ml-172005.iam.gserviceaccount.com" DATAFLOW_MACHINE_TYPE = "n1-standard-4" DATAFLOW_MAX_WORKERS = 4 DATAFLOW_DISK_SIZE_GB = 100 EXAMPLE_GEN_BEAM_ARGS = [ "--runner=DataflowRunner", "--project=" + GOOGLE_CLOUD_PROJECT, "--region=" + GOOGLE_CLOUD_REGION, "--service_account_email=" + DATAFLOW_SERVICE_ACCOUNT, "--machine_type=" + DATAFLOW_MACHINE_TYPE, "--experiments=use_runner_v2", "--max_num_workers=" + str(DATAFLOW_MAX_WORKERS), "--disk_size_gb=" + str(DATAFLOW_DISK_SIZE_GB), ] TRANSFORM_BEAM_ARGS = [ "--runner=DataflowRunner", "--project=" + GOOGLE_CLOUD_PROJECT, "--region=" + GOOGLE_CLOUD_REGION, "--service_account_email=" + DATAFLOW_SERVICE_ACCOUNT, "--machine_type=" + DATAFLOW_MACHINE_TYPE, "--experiments=use_runner_v2", "--max_num_workers=" + str(DATAFLOW_MAX_WORKERS), "--disk_size_gb=" + str(DATAFLOW_DISK_SIZE_GB), "--worker_harness_container_image=" + PIPELINE_IMAGE, ] GCP_AI_PLATFORM_TRAINING_ARGS[vertex_training_const.TRAINING_ARGS_KEY][ "worker_pool_specs" ] = [ { "machine_spec": { "machine_type": "n1-standard-8", "accelerator_type": "NVIDIA_TESLA_V100", "accelerator_count": 1, }, "replica_count": 1, "container_spec": { "image_uri": PIPELINE_IMAGE, }, } ] NUM_PARALLEL_TRIALS = 3 GCP_AI_PLATFORM_TUNER_ARGS = { vertex_const.ENABLE_VERTEX_KEY: True, vertex_const.VERTEX_REGION_KEY: GOOGLE_CLOUD_REGION, vertex_tuner_const.TUNING_ARGS_KEY: { "project": GOOGLE_CLOUD_PROJECT, "job_spec": { "worker_pool_specs": [ { "machine_spec": { "machine_type": "n1-standard-4", "accelerator_type": "NVIDIA_TESLA_K80", "accelerator_count": 1, }, "replica_count": 1, "container_spec": { "image_uri": PIPELINE_IMAGE, }, } ], }, }, vertex_tuner_const.REMOTE_TRIALS_WORKING_DIR_KEY: os.path.join( PIPELINE_ROOT, "trials" ), "use_gpu": True, "hyperparameters": HYPER_PARAMETERS, } GCP_AI_PLATFORM_SERVING_ARGS = { vertex_const.ENABLE_VERTEX_KEY: True, vertex_const.VERTEX_REGION_KEY: GOOGLE_CLOUD_REGION, vertex_const.VERTEX_CONTAINER_IMAGE_URI_KEY: "us-docker.pkg.dev/vertex-ai/prediction/tf2-cpu.2-8:latest", vertex_const.SERVING_ARGS_KEY: { "project_id": GOOGLE_CLOUD_PROJECT, "deployed_model_display_name": PIPELINE_NAME.replace("-", "_"), "endpoint_name": "prediction-" + PIPELINE_NAME.replace("-", "_"), "traffic_split": {"0": 100}, "machine_type": "n1-standard-4", "min_replica_count": 1, "max_replica_count": 1, }, }
deep-diver/mlops-hf-tf-vision-models
advanced_part2/pipeline/kubeflow_pipeline.py
from typing import Any, Dict, List, Optional, Text import tensorflow_model_analysis as tfma from tfx import v1 as tfx from ml_metadata.proto import metadata_store_pb2 from tfx.proto import example_gen_pb2 from tfx.components import ImportExampleGen from tfx.components import StatisticsGen from tfx.components import ExampleValidator from tfx.components import Transform from tfx.components import Evaluator from tfx.extensions.google_cloud_ai_platform.trainer.component import ( Trainer as VertexTrainer, ) from tfx.extensions.google_cloud_ai_platform.pusher.component import ( Pusher as VertexPusher, ) from tfx.extensions.google_cloud_ai_platform.tuner.component import Tuner as VertexTuner from tfx.orchestration import pipeline from tfx.proto import example_gen_pb2 from tfx.proto import tuner_pb2 from tfx.types import Channel from tfx.types.standard_artifacts import Model from tfx.types.standard_artifacts import ModelBlessing from tfx.dsl.components.common import resolver from tfx.dsl.experimental.latest_blessed_model_resolver import ( LatestBlessedModelResolver, ) def create_pipeline( pipeline_name: Text, pipeline_root: Text, data_path: Text, schema_path: Text, modules: Dict[Text, Text], eval_configs: tfma.EvalConfig, metadata_connection_config: Optional[metadata_store_pb2.ConnectionConfig] = None, ai_platform_training_args: Optional[Dict[Text, Text]] = None, ai_platform_tuner_args: Optional[Dict[Text, Text]] = None, tuner_args: tuner_pb2.TuneArgs = None, ai_platform_serving_args: Optional[Dict[Text, Any]] = None, example_gen_beam_args: Optional[List] = None, transform_beam_args: Optional[List] = None, ) -> tfx.dsl.Pipeline: components = [] input_config = example_gen_pb2.Input( splits=[ example_gen_pb2.Input.Split(name="train", pattern="train-*.tfrec"), example_gen_pb2.Input.Split(name="eval", pattern="val-*.tfrec"), ] ) example_gen = ImportExampleGen(input_base=data_path, input_config=input_config) if example_gen_beam_args is not None: example_gen.with_beam_pipeline_args(example_gen_beam_args) components.append(example_gen) statistics_gen = StatisticsGen(examples=example_gen.outputs["examples"]) components.append(statistics_gen) schema_gen = tfx.components.ImportSchemaGen(schema_file=schema_path) components.append(schema_gen) example_validator = ExampleValidator( statistics=statistics_gen.outputs["statistics"], schema=schema_gen.outputs["schema"], ) components.append(example_validator) transform_args = { "examples": example_gen.outputs["examples"], "schema": schema_gen.outputs["schema"], "preprocessing_fn": modules["preprocessing_fn"], } transform = Transform(**transform_args) if transform_beam_args is not None: transform.with_beam_pipeline_args(transform_beam_args) components.append(transform) tuner = VertexTuner( tuner_fn=modules["tuner_fn"], examples=transform.outputs["transformed_examples"], transform_graph=transform.outputs["transform_graph"], tune_args=tuner_args, custom_config=ai_platform_tuner_args, ) components.append(tuner) trainer_args = { "run_fn": modules["training_fn"], "transformed_examples": transform.outputs["transformed_examples"], "transform_graph": transform.outputs["transform_graph"], "schema": schema_gen.outputs["schema"], "hyperparameters": tuner.outputs["best_hyperparameters"], "custom_config": ai_platform_training_args, } trainer = VertexTrainer(**trainer_args) components.append(trainer) model_resolver = resolver.Resolver( strategy_class=LatestBlessedModelResolver, model=Channel(type=Model), model_blessing=Channel(type=ModelBlessing), ).with_id("latest_blessed_model_resolver") components.append(model_resolver) evaluator = Evaluator( examples=example_gen.outputs["examples"], model=trainer.outputs["model"], baseline_model=model_resolver.outputs["model"], eval_config=eval_configs, ) components.append(evaluator) pusher_args = { "model": trainer.outputs["model"], "model_blessing": evaluator.outputs["blessing"], "custom_config": ai_platform_serving_args, } pusher = VertexPusher(**pusher_args) # pylint: disable=unused-variable components.append(pusher) return pipeline.Pipeline( pipeline_name=pipeline_name, pipeline_root=pipeline_root, components=components, enable_cache=True, metadata_connection_config=metadata_connection_config, )
deep-diver/mlops-hf-tf-vision-models
advanced_part2/pipeline/local_pipeline.py
from typing import Dict, Optional, Text import tensorflow_model_analysis as tfma from tfx import v1 as tfx from ml_metadata.proto import metadata_store_pb2 from tfx.proto import example_gen_pb2 from tfx.components import ImportExampleGen from tfx.components import StatisticsGen from tfx.components import ExampleValidator from tfx.components import Transform from tfx.components import Tuner from tfx.components import Trainer from tfx.components import Evaluator from tfx.components import Pusher from tfx.orchestration import pipeline from tfx.proto import example_gen_pb2 from tfx.types import Channel from tfx.types.standard_artifacts import Model from tfx.types.standard_artifacts import ModelBlessing from tfx.dsl.components.common import resolver from tfx.dsl.experimental.latest_blessed_model_resolver import ( LatestBlessedModelResolver, ) def create_pipeline( pipeline_name: Text, pipeline_root: Text, data_path: Text, schema_path: Text, modules: Dict[Text, Text], hyperparameters: Dict[Text, Text], eval_configs: tfma.EvalConfig, serving_model_dir: Text, metadata_connection_config: Optional[metadata_store_pb2.ConnectionConfig] = None, ) -> tfx.dsl.Pipeline: components = [] input_config = example_gen_pb2.Input( splits=[ example_gen_pb2.Input.Split(name="train", pattern="train-00-*.tfrec"), example_gen_pb2.Input.Split(name="eval", pattern="val-00-*.tfrec"), ] ) example_gen = ImportExampleGen(input_base=data_path, input_config=input_config) components.append(example_gen) statistics_gen = StatisticsGen(examples=example_gen.outputs["examples"]) components.append(statistics_gen) schema_gen = tfx.components.ImportSchemaGen(schema_file=schema_path) components.append(schema_gen) example_validator = ExampleValidator( statistics=statistics_gen.outputs["statistics"], schema=schema_gen.outputs["schema"], ) components.append(example_validator) transform_args = { "examples": example_gen.outputs["examples"], "schema": schema_gen.outputs["schema"], "preprocessing_fn": modules["preprocessing_fn"], } transform = Transform(**transform_args) components.append(transform) tuner = Tuner( tuner_fn=modules["tuner_fn"], examples=transform.outputs["transformed_examples"], schema=schema_gen.outputs["schema"], transform_graph=transform.outputs["transform_graph"], custom_config={"hyperparameters": hyperparameters}, ) components.append(tuner) trainer_args = { "run_fn": modules["training_fn"], "transformed_examples": transform.outputs["transformed_examples"], "transform_graph": transform.outputs["transform_graph"], "schema": schema_gen.outputs["schema"], "hyperparameters": tuner.outputs["best_hyperparameters"], "custom_config": {"is_local": True}, } trainer = Trainer(**trainer_args) components.append(trainer) model_resolver = resolver.Resolver( strategy_class=LatestBlessedModelResolver, model=Channel(type=Model), model_blessing=Channel(type=ModelBlessing), ).with_id("latest_blessed_model_resolver") components.append(model_resolver) evaluator = Evaluator( examples=example_gen.outputs["examples"], model=trainer.outputs["model"], baseline_model=model_resolver.outputs["model"], eval_config=eval_configs, ) components.append(evaluator) pusher_args = { "model": trainer.outputs["model"], "model_blessing": evaluator.outputs["blessing"], "push_destination": tfx.proto.PushDestination( filesystem=tfx.proto.PushDestination.Filesystem( base_directory=serving_model_dir ) ), } pusher = Pusher(**pusher_args) # pylint: disable=unused-variable components.append(pusher) return pipeline.Pipeline( pipeline_name=pipeline_name, pipeline_root=pipeline_root, components=components, enable_cache=False, metadata_connection_config=metadata_connection_config, )
deep-diver/mlops-hf-tf-vision-models
basic/modules/ViT.py
import tensorflow as tf from transformers import TFViTForImageClassification from .common import LABELS from .common import PRETRAIN_CHECKPOINT from .utils import INFO def build_model(): id2label = {str(i): c for i, c in enumerate(LABELS)} label2id = {c: str(i) for i, c in enumerate(LABELS)} model = TFViTForImageClassification.from_pretrained( PRETRAIN_CHECKPOINT, num_labels=len(LABELS), label2id=label2id, id2label=id2label, ) model.layers[0].trainable = False loss = tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True) model.compile(optimizer="adam", loss=loss, metrics=["accuracy"]) INFO(model.summary()) return model
deep-diver/mlops-hf-tf-vision-models
basic/modules/common.py
IMAGE_TFREC_KEY = "image" IMAGE_SHAPE_TFREC_KEY = "image_shape" LABEL_TFREC_KEY = "label" IMAGE_MODEL_KEY = "pixel_values" LABEL_MODEL_KEY = "labels" CONCRETE_INPUT = "pixel_values" PRETRAIN_CHECKPOINT = "google/vit-base-patch16-224-in21k" LABELS = ["angular_leaf_spot", "bean_rust", "healthy"]
deep-diver/mlops-hf-tf-vision-models
basic/modules/hyperparams.py
EPOCHS = 1 BATCH_SIZE = 32 TRAIN_BATCH_SIZE = 32 EVAL_BATCH_SIZE = 32 TRAIN_LENGTH = 1034 EVAL_LENGTH = 128
deep-diver/mlops-hf-tf-vision-models
basic/modules/signatures.py
import tensorflow as tf from transformers import ViTFeatureExtractor from .common import PRETRAIN_CHECKPOINT from .common import CONCRETE_INPUT feature_extractor = ViTFeatureExtractor.from_pretrained(PRETRAIN_CHECKPOINT) def _normalize_img( img, mean=feature_extractor.image_mean, std=feature_extractor.image_std ): img = img / 255 mean = tf.constant(mean) std = tf.constant(std) return (img - mean) / std def _preprocess_serving(string_input): decoded_input = tf.io.decode_base64(string_input) decoded = tf.io.decode_jpeg(decoded_input, channels=3) resized = tf.image.resize(decoded, size=(224, 224)) normalized = _normalize_img(resized) normalized = tf.transpose( normalized, (2, 0, 1) ) # Since HF models are channel-first. return normalized @tf.function(input_signature=[tf.TensorSpec([None], tf.string)]) def _preprocess_fn(string_input): decoded_images = tf.map_fn( _preprocess_serving, string_input, dtype=tf.float32, back_prop=False ) return {CONCRETE_INPUT: decoded_images} def model_exporter(model: tf.keras.Model): m_call = tf.function(model.call).get_concrete_function( tf.TensorSpec(shape=[None, 3, 224, 224], dtype=tf.float32, name=CONCRETE_INPUT) ) @tf.function(input_signature=[tf.TensorSpec([None], tf.string)]) def serving_fn(string_input): labels = tf.constant(list(model.config.id2label.values()), dtype=tf.string) images = _preprocess_fn(string_input) predictions = m_call(**images) indices = tf.argmax(predictions.logits, axis=1) pred_source = tf.gather(params=labels, indices=indices) probs = tf.nn.softmax(predictions.logits, axis=1) pred_confidence = tf.reduce_max(probs, axis=1) return {"label": pred_source, "confidence": pred_confidence} return serving_fn
deep-diver/mlops-hf-tf-vision-models
basic/modules/train.py
from tfx.components.trainer.fn_args_utils import FnArgs from .train_data import input_fn from .ViT import build_model from .signatures import model_exporter from .hyperparams import TRAIN_BATCH_SIZE, EVAL_BATCH_SIZE from .hyperparams import TRAIN_LENGTH, EVAL_LENGTH from .hyperparams import EPOCHS def run_fn(fn_args: FnArgs): train_dataset = input_fn( fn_args.train_files, is_train=True, batch_size=TRAIN_BATCH_SIZE, ) eval_dataset = input_fn( fn_args.eval_files, is_train=False, batch_size=EVAL_BATCH_SIZE, ) model = build_model() model.fit( train_dataset, steps_per_epoch=TRAIN_LENGTH // TRAIN_BATCH_SIZE, validation_data=eval_dataset, validation_steps=EVAL_LENGTH // TRAIN_BATCH_SIZE, epochs=EPOCHS, ) model.save( fn_args.serving_model_dir, save_format="tf", signatures=model_exporter(model) )
deep-diver/mlops-hf-tf-vision-models
basic/modules/train_data.py
from typing import List import tensorflow as tf from .utils import INFO from .common import IMAGE_TFREC_KEY, IMAGE_SHAPE_TFREC_KEY, LABEL_TFREC_KEY from .common import IMAGE_MODEL_KEY, LABEL_MODEL_KEY from .hyperparams import BATCH_SIZE def _parse_tfr(proto): feature_description = { IMAGE_TFREC_KEY: tf.io.VarLenFeature(tf.float32), IMAGE_SHAPE_TFREC_KEY: tf.io.VarLenFeature(tf.int64), LABEL_TFREC_KEY: tf.io.VarLenFeature(tf.int64), } rec = tf.io.parse_single_example(proto, feature_description) image_shape = tf.sparse.to_dense(rec[IMAGE_SHAPE_TFREC_KEY]) image = tf.reshape(tf.sparse.to_dense(rec[IMAGE_TFREC_KEY]), image_shape) label = tf.sparse.to_dense(rec[LABEL_TFREC_KEY]) return {IMAGE_MODEL_KEY: image, LABEL_MODEL_KEY: label} def _preprocess(example_batch): images = example_batch[IMAGE_MODEL_KEY] images = tf.transpose( images, perm=[0, 1, 2, 3] ) # (batch_size, height, width, num_channels) images = tf.image.resize(images, (224, 224)) images = tf.transpose(images, perm=[0, 3, 1, 2]) labels = example_batch[LABEL_MODEL_KEY] labels = tf.transpose(labels, perm=[0, 1]) # So, that TF can evaluation the shapes. return {IMAGE_MODEL_KEY: images, LABEL_MODEL_KEY: labels} def input_fn( file_pattern: List[str], batch_size: int = BATCH_SIZE, is_train: bool = False, ) -> tf.data.Dataset: INFO(f"Reading data from: {file_pattern}") dataset = tf.data.TFRecordDataset( tf.io.gfile.glob(file_pattern[0] + ".gz"), num_parallel_reads=tf.data.AUTOTUNE, compression_type="GZIP", ).map(_parse_tfr, num_parallel_calls=tf.data.AUTOTUNE) if is_train: dataset = dataset.shuffle(batch_size * 2) dataset = dataset.batch(batch_size) dataset = dataset.prefetch(tf.data.AUTOTUNE) dataset = dataset.map(_preprocess) return dataset
deep-diver/mlops-hf-tf-vision-models
basic/modules/utils.py
import absl def INFO(text: str): absl.logging.info(text)
deep-diver/mlops-hf-tf-vision-models
basic/pipeline/configs.py
import os # pylint: disable=unused-import import tfx.extensions.google_cloud_ai_platform.constants as vertex_const import tfx.extensions.google_cloud_ai_platform.trainer.executor as vertex_training_const PIPELINE_NAME = "vit-e2e-pipeline-basic" try: import google.auth # pylint: disable=g-import-not-at-top # pytype: disable=import-error try: _, GOOGLE_CLOUD_PROJECT = google.auth.default() except google.auth.exceptions.DefaultCredentialsError: GOOGLE_CLOUD_PROJECT = "gcp-ml-172005" except ImportError: GOOGLE_CLOUD_PROJECT = "gcp-ml-172005" GOOGLE_CLOUD_REGION = "us-central1" GCS_BUCKET_NAME = GOOGLE_CLOUD_PROJECT + "-complete-mlops" PIPELINE_IMAGE = f"gcr.io/{GOOGLE_CLOUD_PROJECT}/{PIPELINE_NAME}" OUTPUT_DIR = os.path.join("gs://", GCS_BUCKET_NAME) PIPELINE_ROOT = os.path.join(OUTPUT_DIR, "tfx_pipeline_output", PIPELINE_NAME) DATA_PATH = "gs://beans-lowres/tfrecords/" TRAINING_FN = "modules.train.run_fn" EXAMPLE_GEN_BEAM_ARGS = None GCP_AI_PLATFORM_TRAINING_ARGS = { vertex_const.ENABLE_VERTEX_KEY: True, vertex_const.VERTEX_REGION_KEY: GOOGLE_CLOUD_REGION, vertex_training_const.TRAINING_ARGS_KEY: { "project": GOOGLE_CLOUD_PROJECT, "worker_pool_specs": [ { "machine_spec": { "machine_type": "n1-standard-4", "accelerator_type": "NVIDIA_TESLA_K80", "accelerator_count": 1, }, "replica_count": 1, "container_spec": { "image_uri": PIPELINE_IMAGE, }, } ], }, "use_gpu": True, } fullres_data = os.environ.get("FULL_RES_DATA", "false") if fullres_data.lower() == "true": DATA_PATH = "gs://beans-fullres/tfrecords/" DATAFLOW_SERVICE_ACCOUNT = "csp-gde-dataflow@gcp-ml-172005.iam.gserviceaccount.com" DATAFLOW_MACHINE_TYPE = "n1-standard-4" DATAFLOW_MAX_WORKERS = 4 DATAFLOW_DISK_SIZE_GB = 100 EXAMPLE_GEN_BEAM_ARGS = [ "--runner=DataflowRunner", "--project=" + GOOGLE_CLOUD_PROJECT, "--region=" + GOOGLE_CLOUD_REGION, "--service_account_email=" + DATAFLOW_SERVICE_ACCOUNT, "--machine_type=" + DATAFLOW_MACHINE_TYPE, "--experiments=use_runner_v2", "--max_num_workers=" + str(DATAFLOW_MAX_WORKERS), "--disk_size_gb=" + str(DATAFLOW_DISK_SIZE_GB), ] GCP_AI_PLATFORM_TRAINING_ARGS[vertex_training_const.TRAINING_ARGS_KEY][ "worker_pool_specs" ] = [ { "machine_spec": { "machine_type": "n1-standard-8", "accelerator_type": "NVIDIA_TESLA_V100", "accelerator_count": 1, }, "replica_count": 1, "container_spec": { "image_uri": PIPELINE_IMAGE, }, } ] GCP_AI_PLATFORM_SERVING_ARGS = { vertex_const.ENABLE_VERTEX_KEY: True, vertex_const.VERTEX_REGION_KEY: GOOGLE_CLOUD_REGION, vertex_const.VERTEX_CONTAINER_IMAGE_URI_KEY: "us-docker.pkg.dev/vertex-ai/prediction/tf2-cpu.2-8:latest", vertex_const.SERVING_ARGS_KEY: { "project_id": GOOGLE_CLOUD_PROJECT, "deployed_model_display_name": PIPELINE_NAME.replace("-", "_"), "endpoint_name": "prediction-" + PIPELINE_NAME.replace("-", "_"), "traffic_split": {"0": 100}, "machine_type": "n1-standard-4", "min_replica_count": 1, "max_replica_count": 1, }, }
deep-diver/mlops-hf-tf-vision-models
basic/pipeline/kubeflow_pipeline.py
from typing import Any, Dict, List, Optional, Text from tfx import v1 as tfx from ml_metadata.proto import metadata_store_pb2 from tfx.proto import example_gen_pb2 from tfx.components import ImportExampleGen from tfx.extensions.google_cloud_ai_platform.trainer.component import ( Trainer as VertexTrainer, ) from tfx.extensions.google_cloud_ai_platform.pusher.component import ( Pusher as VertexPusher, ) from tfx.orchestration import pipeline from tfx.proto import example_gen_pb2 def create_pipeline( pipeline_name: Text, pipeline_root: Text, data_path: Text, modules: Dict[Text, Text], metadata_connection_config: Optional[metadata_store_pb2.ConnectionConfig] = None, ai_platform_training_args: Optional[Dict[Text, Text]] = None, ai_platform_serving_args: Optional[Dict[Text, Any]] = None, example_gen_beam_args: Optional[List] = None, ) -> tfx.dsl.Pipeline: components = [] input_config = example_gen_pb2.Input( splits=[ example_gen_pb2.Input.Split(name="train", pattern="train-*.tfrec"), example_gen_pb2.Input.Split(name="eval", pattern="val-*.tfrec"), ] ) example_gen = ImportExampleGen(input_base=data_path, input_config=input_config) if example_gen_beam_args is not None: example_gen.with_beam_pipeline_args(example_gen_beam_args) components.append(example_gen) trainer_args = { "run_fn": modules["training_fn"], "examples": example_gen.outputs["examples"], "custom_config": ai_platform_training_args, } trainer = VertexTrainer(**trainer_args) components.append(trainer) pusher_args = { "model": trainer.outputs["model"], "custom_config": ai_platform_serving_args, } pusher = VertexPusher(**pusher_args) # pylint: disable=unused-variable components.append(pusher) return pipeline.Pipeline( pipeline_name=pipeline_name, pipeline_root=pipeline_root, components=components, enable_cache=True, metadata_connection_config=metadata_connection_config, )
deep-diver/mlops-hf-tf-vision-models
basic/pipeline/local_pipeline.py
from typing import Dict, Optional, Text from tfx import v1 as tfx from ml_metadata.proto import metadata_store_pb2 from tfx.proto import example_gen_pb2 from tfx.components import ImportExampleGen from tfx.components import Trainer from tfx.components import Pusher from tfx.orchestration import pipeline from tfx.proto import example_gen_pb2 def create_pipeline( pipeline_name: Text, pipeline_root: Text, data_path: Text, modules: Dict[Text, Text], serving_model_dir: Text, metadata_connection_config: Optional[metadata_store_pb2.ConnectionConfig] = None, ) -> tfx.dsl.Pipeline: components = [] input_config = example_gen_pb2.Input( splits=[ example_gen_pb2.Input.Split(name="train", pattern="train-00-*.tfrec"), example_gen_pb2.Input.Split(name="eval", pattern="val-00-*.tfrec"), ] ) example_gen = ImportExampleGen(input_base=data_path, input_config=input_config) components.append(example_gen) trainer = Trainer( run_fn=modules["training_fn"], examples=example_gen.outputs["examples"], ) components.append(trainer) pusher_args = { "model": trainer.outputs["model"], "push_destination": tfx.proto.PushDestination( filesystem=tfx.proto.PushDestination.Filesystem( base_directory=serving_model_dir ) ), } pusher = Pusher(**pusher_args) # pylint: disable=unused-variable components.append(pusher) return pipeline.Pipeline( pipeline_name=pipeline_name, pipeline_root=pipeline_root, components=components, enable_cache=False, metadata_connection_config=metadata_connection_config, )
deep-diver/mlops-hf-tf-vision-models
hf_integration/modules/ViT.py
import tensorflow as tf import keras_tuner from transformers import TFViTForImageClassification from .common import LABELS from .common import PRETRAIN_CHECKPOINT from .utils import INFO def build_model(hparams: keras_tuner.HyperParameters): id2label = {str(i): c for i, c in enumerate(LABELS)} label2id = {c: str(i) for i, c in enumerate(LABELS)} model = TFViTForImageClassification.from_pretrained( PRETRAIN_CHECKPOINT, num_labels=len(LABELS), label2id=label2id, id2label=id2label, ) model.layers[0].trainable = False loss = tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True) optimizer = tf.keras.optimizers.Adam(learning_rate=hparams.get("learning_rate")) model.compile(optimizer=optimizer, loss=loss, metrics=["accuracy"]) INFO(model.summary()) return model
deep-diver/mlops-hf-tf-vision-models
hf_integration/modules/common.py
IMAGE_TFREC_KEY = "image" IMAGE_SHAPE_TFREC_KEY = "image_shape" LABEL_TFREC_KEY = "label" MODEL_INPUT_IMAGE_KEY = "pixel_values" MODEL_INPUT_LABEL_KEY = "labels" IMAGE_MODEL_KEY = "pixel_values" LABEL_MODEL_KEY = "labels" CONCRETE_INPUT = "pixel_values" PRETRAIN_CHECKPOINT = "google/vit-base-patch16-224-in21k" LABELS = ["angular_leaf_spot", "bean_rust", "healthy"]
deep-diver/mlops-hf-tf-vision-models
hf_integration/modules/hyperparams.py
import keras_tuner EPOCHS = 10 BATCH_SIZE = 32 TRAIN_BATCH_SIZE = 32 EVAL_BATCH_SIZE = 32 TRAIN_LENGTH = 1034 EVAL_LENGTH = 128 INPUT_IMG_SIZE = 224 def get_hyperparameters(hyperparameters) -> keras_tuner.HyperParameters: hp_set = keras_tuner.HyperParameters() for hp in hyperparameters: hp_set.Choice( hp, hyperparameters[hp]["values"], default=hyperparameters[hp]["default"] ) return hp_set
deep-diver/mlops-hf-tf-vision-models
hf_integration/modules/preprocessing.py
import tensorflow as tf from .common import IMAGE_TFREC_KEY, LABEL_TFREC_KEY from .common import IMAGE_MODEL_KEY, LABEL_MODEL_KEY from .hyperparams import INPUT_IMG_SIZE def preprocessing_fn(inputs): """tf.transform's callback function for preprocessing inputs. Args: inputs: map from feature keys to raw not-yet-transformed features. Returns: Map from string feature key to transformed feature operations. """ # print(inputs) outputs = {} inputs[IMAGE_TFREC_KEY] = tf.image.resize( inputs[IMAGE_TFREC_KEY], [INPUT_IMG_SIZE, INPUT_IMG_SIZE] ) inputs[IMAGE_TFREC_KEY] = inputs[IMAGE_TFREC_KEY] / 255.0 inputs[IMAGE_TFREC_KEY] = tf.transpose(inputs[IMAGE_TFREC_KEY], [0, 3, 1, 2]) outputs[IMAGE_MODEL_KEY] = inputs[IMAGE_TFREC_KEY] outputs[LABEL_MODEL_KEY] = inputs[LABEL_TFREC_KEY] return outputs
deep-diver/mlops-hf-tf-vision-models
hf_integration/modules/signatures.py
from typing import Dict import tensorflow as tf import tensorflow_transform as tft from transformers import ViTFeatureExtractor from .common import PRETRAIN_CHECKPOINT from .common import CONCRETE_INPUT from .common import LABEL_MODEL_KEY feature_extractor = ViTFeatureExtractor.from_pretrained(PRETRAIN_CHECKPOINT) def _normalize_img( img, mean=feature_extractor.image_mean, std=feature_extractor.image_std ): img = img / 255 mean = tf.constant(mean) std = tf.constant(std) return (img - mean) / std def _preprocess_serving(string_input): decoded_input = tf.io.decode_base64(string_input) decoded = tf.io.decode_jpeg(decoded_input, channels=3) resized = tf.image.resize(decoded, size=(224, 224)) normalized = _normalize_img(resized) normalized = tf.transpose( normalized, (2, 0, 1) ) # Since HF models are channel-first. return normalized @tf.function(input_signature=[tf.TensorSpec([None], tf.string)]) def _preprocess_fn(string_input): decoded_images = tf.map_fn( _preprocess_serving, string_input, dtype=tf.float32, back_prop=False ) return {CONCRETE_INPUT: decoded_images} def model_exporter(model: tf.keras.Model): m_call = tf.function(model.call).get_concrete_function( tf.TensorSpec(shape=[None, 3, 224, 224], dtype=tf.float32, name=CONCRETE_INPUT) ) @tf.function(input_signature=[tf.TensorSpec([None], tf.string)]) def serving_fn(string_input): labels = tf.constant(list(model.config.id2label.values()), dtype=tf.string) images = _preprocess_fn(string_input) predictions = m_call(**images) indices = tf.argmax(predictions.logits, axis=1) pred_source = tf.gather(params=labels, indices=indices) probs = tf.nn.softmax(predictions.logits, axis=1) pred_confidence = tf.reduce_max(probs, axis=1) return {"label": pred_source, "confidence": pred_confidence} return serving_fn def transform_features_signature( model: tf.keras.Model, tf_transform_output: tft.TFTransformOutput ): """ transform_features_signature simply returns a function that transforms any data of the type of tf.Example which is denoted as the type of sta ndard_artifacts.Examples in TFX. The purpose of this function is to ap ply Transform Graph obtained from Transform component to the data prod uced by ImportExampleGen. This function will be used in the Evaluator component, so the raw evaluation inputs from ImportExampleGen can be a pporiately transformed that the model could understand. """ # basically, what Transform component emits is a SavedModel that knows # how to transform data. transform_features_layer() simply returns the # layer from the Transform. model.tft_layer = tf_transform_output.transform_features_layer() @tf.function( input_signature=[tf.TensorSpec(shape=[None], dtype=tf.string, name="examples")] ) def serve_tf_examples_fn(serialized_tf_examples): """ raw_feature_spec returns a set of feature maps(dict) for the input TFRecords based on the knowledge that Transform component has lear ned(learn doesn't mean training here). By using this information, the raw data from ImportExampleGen could be parsed with tf.io.parse _example utility function. Then, it is passed to the model.tft_layer, so the final output we get is the transformed data of the raw input. """ feature_spec = tf_transform_output.raw_feature_spec() parsed_features = tf.io.parse_example(serialized_tf_examples, feature_spec) transformed_features = model.tft_layer(parsed_features) return transformed_features return serve_tf_examples_fn def tf_examples_serving_signature(model, tf_transform_output): """ tf_examples_serving_signature simply returns a function that performs data transformation(preprocessing) and model prediction in a sequential manner. How data transformation is done is idential to the process of transform_features_signature function. """ model.tft_layer = tf_transform_output.transform_features_layer() @tf.function( input_signature=[tf.TensorSpec(shape=[None], dtype=tf.string, name="examples")] ) def serve_tf_examples_fn( serialized_tf_example: tf.Tensor, ) -> Dict[str, tf.Tensor]: raw_feature_spec = tf_transform_output.raw_feature_spec() raw_features = tf.io.parse_example(serialized_tf_example, raw_feature_spec) transformed_features = model.tft_layer(raw_features) logits = model(transformed_features).logits return {LABEL_MODEL_KEY: logits} return serve_tf_examples_fn
deep-diver/mlops-hf-tf-vision-models
hf_integration/modules/train.py
import keras_tuner import tensorflow_transform as tft from tfx.components.trainer.fn_args_utils import FnArgs from .train_data import input_fn from .ViT import build_model from .signatures import ( model_exporter, transform_features_signature, tf_examples_serving_signature, ) from .hyperparams import TRAIN_BATCH_SIZE, EVAL_BATCH_SIZE from .hyperparams import TRAIN_LENGTH, EVAL_LENGTH from .hyperparams import EPOCHS from .utils import INFO def run_fn(fn_args: FnArgs): custom_config = fn_args.custom_config epochs = EPOCHS if custom_config is not None: if "is_local" in custom_config: epochs = 1 tf_transform_output = tft.TFTransformOutput(fn_args.transform_output) train_dataset = input_fn( fn_args.train_files, fn_args.data_accessor, tf_transform_output, is_train=True, batch_size=TRAIN_BATCH_SIZE, ) eval_dataset = input_fn( fn_args.eval_files, fn_args.data_accessor, tf_transform_output, is_train=False, batch_size=EVAL_BATCH_SIZE, ) hparams = keras_tuner.HyperParameters.from_config(fn_args.hyperparameters) INFO(f"HyperParameters for training: {hparams.get_config()}") model = build_model(hparams) model.fit( train_dataset, steps_per_epoch=TRAIN_LENGTH // TRAIN_BATCH_SIZE, validation_data=eval_dataset, validation_steps=EVAL_LENGTH // TRAIN_BATCH_SIZE, epochs=epochs, ) model.save( fn_args.serving_model_dir, save_format="tf", signatures={ "serving_default": model_exporter(model), "transform_features": transform_features_signature( model, tf_transform_output ), "from_examples": tf_examples_serving_signature(model, tf_transform_output), }, )
deep-diver/mlops-hf-tf-vision-models
hf_integration/modules/train_data.py
from typing import List import tensorflow as tf import tensorflow_transform as tft from tfx_bsl.tfxio import dataset_options from tfx.components.trainer.fn_args_utils import DataAccessor from .utils import INFO from .common import LABEL_MODEL_KEY from .hyperparams import BATCH_SIZE def input_fn( file_pattern: List[str], data_accessor: DataAccessor, tf_transform_output: tft.TFTransformOutput, is_train: bool = False, batch_size: int = BATCH_SIZE, ) -> tf.data.Dataset: INFO(f"Reading data from: {file_pattern}") dataset = data_accessor.tf_dataset_factory( file_pattern, dataset_options.TensorFlowDatasetOptions( batch_size=batch_size, label_key=LABEL_MODEL_KEY ), tf_transform_output.transformed_metadata.schema, ) return dataset
deep-diver/mlops-hf-tf-vision-models
hf_integration/modules/tuning.py
import keras_tuner import tensorflow_transform as tft from tfx.components.trainer.fn_args_utils import FnArgs from tfx.v1.components import TunerFnResult from .train_data import input_fn from .ViT import build_model from .hyperparams import TRAIN_BATCH_SIZE, EVAL_BATCH_SIZE from .hyperparams import TRAIN_LENGTH, EVAL_LENGTH from .hyperparams import get_hyperparameters def tuner_fn(fn_args: FnArgs) -> TunerFnResult: hyperparameters = fn_args.custom_config["hyperparameters"] tuner = keras_tuner.RandomSearch( build_model, max_trials=6, hyperparameters=get_hyperparameters(hyperparameters), allow_new_entries=False, objective=keras_tuner.Objective("val_accuracy", "max"), directory=fn_args.working_dir, project_name="ViT MLOps Advanced Part2", ) tf_transform_output = tft.TFTransformOutput(fn_args.transform_graph_path) train_dataset = input_fn( fn_args.train_files, fn_args.data_accessor, tf_transform_output, is_train=True, batch_size=TRAIN_BATCH_SIZE, ) eval_dataset = input_fn( fn_args.eval_files, fn_args.data_accessor, tf_transform_output, is_train=False, batch_size=EVAL_BATCH_SIZE, ) return TunerFnResult( tuner=tuner, fit_kwargs={ "x": train_dataset, "validation_data": eval_dataset, "steps_per_epoch": TRAIN_LENGTH // TRAIN_BATCH_SIZE, "validation_steps": EVAL_LENGTH // EVAL_BATCH_SIZE, }, )
deep-diver/mlops-hf-tf-vision-models
hf_integration/modules/utils.py
import absl def INFO(text: str): absl.logging.info(text)
deep-diver/mlops-hf-tf-vision-models
hf_integration/pipeline/configs.py
import os import tensorflow_model_analysis as tfma import tfx.extensions.google_cloud_ai_platform.constants as vertex_const import tfx.extensions.google_cloud_ai_platform.trainer.executor as vertex_training_const import tfx.extensions.google_cloud_ai_platform.tuner.executor as vertex_tuner_const PIPELINE_NAME = "vit-e2e-pipeline-hf-integration" try: import google.auth # pylint: disable=g-import-not-at-top # pytype: disable=import-error try: _, GOOGLE_CLOUD_PROJECT = google.auth.default() except google.auth.exceptions.DefaultCredentialsError: GOOGLE_CLOUD_PROJECT = "gcp-ml-172005" except ImportError: GOOGLE_CLOUD_PROJECT = "gcp-ml-172005" GOOGLE_CLOUD_REGION = "us-central1" GCS_BUCKET_NAME = GOOGLE_CLOUD_PROJECT + "-complete-mlops" PIPELINE_IMAGE = f"gcr.io/{GOOGLE_CLOUD_PROJECT}/{PIPELINE_NAME}" OUTPUT_DIR = os.path.join("gs://", GCS_BUCKET_NAME) PIPELINE_ROOT = os.path.join(OUTPUT_DIR, "tfx_pipeline_output", PIPELINE_NAME) DATA_PATH = "gs://beans-lowres/tfrecords/" SCHEMA_PATH = "pipeline/schema.pbtxt" TRAINING_FN = "modules.train.run_fn" TUNER_FN = "modules.tuning.tuner_fn" PREPROCESSING_FN = "modules.preprocessing.preprocessing_fn" EXAMPLE_GEN_BEAM_ARGS = None TRANSFORM_BEAM_ARGS = None TRAIN_BATCH_SIZE = 32 EVAL_BATCH_SIZE = 32 TRAIN_LENGTH = 1034 EVAL_LENGTH = 128 HYPER_PARAMETERS = { "learning_rate": {"values": [1e-4, 1e-3, 1e-2, 1e-1], "default": 1e-4}, } EVAL_CONFIGS = tfma.EvalConfig( model_specs=[ tfma.ModelSpec( signature_name="from_examples", preprocessing_function_names=["transform_features"], label_key="labels", prediction_key="labels", ) ], slicing_specs=[tfma.SlicingSpec()], metrics_specs=[ tfma.MetricsSpec( metrics=[ tfma.MetricConfig( class_name="SparseCategoricalAccuracy", threshold=tfma.MetricThreshold( value_threshold=tfma.GenericValueThreshold( lower_bound={"value": 0.55} ), # Change threshold will be ignored if there is no # baseline model resolved from MLMD (first run). change_threshold=tfma.GenericChangeThreshold( direction=tfma.MetricDirection.HIGHER_IS_BETTER, absolute={"value": -1e-3}, ), ), ) ] ) ], ) GCP_AI_PLATFORM_TRAINING_ARGS = { vertex_const.ENABLE_VERTEX_KEY: True, vertex_const.VERTEX_REGION_KEY: GOOGLE_CLOUD_REGION, vertex_training_const.TRAINING_ARGS_KEY: { "project": GOOGLE_CLOUD_PROJECT, "worker_pool_specs": [ { "machine_spec": { "machine_type": "n1-standard-4", "accelerator_type": "NVIDIA_TESLA_K80", "accelerator_count": 1, }, "replica_count": 1, "container_spec": { "image_uri": PIPELINE_IMAGE, }, } ], }, "use_gpu": True, } fullres_data = os.environ.get("FULL_RES_DATA", "false") if fullres_data.lower() == "true": DATA_PATH = "gs://beans-fullres/tfrecords/" DATAFLOW_SERVICE_ACCOUNT = "csp-gde-dataflow@gcp-ml-172005.iam.gserviceaccount.com" DATAFLOW_MACHINE_TYPE = "n1-standard-4" DATAFLOW_MAX_WORKERS = 4 DATAFLOW_DISK_SIZE_GB = 100 EXAMPLE_GEN_BEAM_ARGS = [ "--runner=DataflowRunner", "--project=" + GOOGLE_CLOUD_PROJECT, "--region=" + GOOGLE_CLOUD_REGION, "--service_account_email=" + DATAFLOW_SERVICE_ACCOUNT, "--machine_type=" + DATAFLOW_MACHINE_TYPE, "--experiments=use_runner_v2", "--max_num_workers=" + str(DATAFLOW_MAX_WORKERS), "--disk_size_gb=" + str(DATAFLOW_DISK_SIZE_GB), ] TRANSFORM_BEAM_ARGS = [ "--runner=DataflowRunner", "--project=" + GOOGLE_CLOUD_PROJECT, "--region=" + GOOGLE_CLOUD_REGION, "--service_account_email=" + DATAFLOW_SERVICE_ACCOUNT, "--machine_type=" + DATAFLOW_MACHINE_TYPE, "--experiments=use_runner_v2", "--max_num_workers=" + str(DATAFLOW_MAX_WORKERS), "--disk_size_gb=" + str(DATAFLOW_DISK_SIZE_GB), "--worker_harness_container_image=" + PIPELINE_IMAGE, ] GCP_AI_PLATFORM_TRAINING_ARGS[vertex_training_const.TRAINING_ARGS_KEY][ "worker_pool_specs" ] = [ { "machine_spec": { "machine_type": "n1-standard-8", "accelerator_type": "NVIDIA_TESLA_V100", "accelerator_count": 1, }, "replica_count": 1, "container_spec": { "image_uri": PIPELINE_IMAGE, }, } ] NUM_PARALLEL_TRIALS = 3 GCP_AI_PLATFORM_TUNER_ARGS = { vertex_const.ENABLE_VERTEX_KEY: True, vertex_const.VERTEX_REGION_KEY: GOOGLE_CLOUD_REGION, vertex_tuner_const.TUNING_ARGS_KEY: { "project": GOOGLE_CLOUD_PROJECT, "job_spec": { "worker_pool_specs": [ { "machine_spec": { "machine_type": "n1-standard-4", "accelerator_type": "NVIDIA_TESLA_K80", "accelerator_count": 1, }, "replica_count": 1, "container_spec": { "image_uri": PIPELINE_IMAGE, }, } ], }, }, vertex_tuner_const.REMOTE_TRIALS_WORKING_DIR_KEY: os.path.join( PIPELINE_ROOT, "trials" ), "use_gpu": True, "hyperparameters": HYPER_PARAMETERS, } GCP_AI_PLATFORM_SERVING_ARGS = { vertex_const.ENABLE_VERTEX_KEY: True, vertex_const.VERTEX_REGION_KEY: GOOGLE_CLOUD_REGION, vertex_const.VERTEX_CONTAINER_IMAGE_URI_KEY: "us-docker.pkg.dev/vertex-ai/prediction/tf2-cpu.2-8:latest", vertex_const.SERVING_ARGS_KEY: { "project_id": GOOGLE_CLOUD_PROJECT, "deployed_model_display_name": PIPELINE_NAME.replace("-", "_"), "endpoint_name": "prediction-" + PIPELINE_NAME.replace("-", "_"), "traffic_split": {"0": 100}, "machine_type": "n1-standard-4", "min_replica_count": 1, "max_replica_count": 1, }, } HF_PUSHER_ARGS = { "username": "chansung", "access_token": "$HF_ACCESS_TOKEN", "repo_name": PIPELINE_NAME, "space_config": { "app_path": "app.gradio", } }
deep-diver/mlops-hf-tf-vision-models
hf_integration/pipeline/kubeflow_pipeline.py
from typing import Any, Dict, List, Optional, Text import tensorflow_model_analysis as tfma from tfx import v1 as tfx from ml_metadata.proto import metadata_store_pb2 from tfx.proto import example_gen_pb2 from tfx.components import ImportExampleGen from tfx.components import StatisticsGen from tfx.components import ExampleValidator from tfx.components import Transform from tfx.components import Evaluator from tfx.extensions.google_cloud_ai_platform.trainer.component import ( Trainer as VertexTrainer, ) from tfx.extensions.google_cloud_ai_platform.pusher.component import ( Pusher as VertexPusher, ) from tfx.extensions.google_cloud_ai_platform.tuner.component import Tuner as VertexTuner from tfx.orchestration import pipeline from tfx.proto import example_gen_pb2 from tfx.proto import tuner_pb2 from tfx.types import Channel from tfx.types.standard_artifacts import Model from tfx.types.standard_artifacts import ModelBlessing from tfx.dsl.components.common import resolver from tfx.dsl.experimental.latest_blessed_model_resolver import ( LatestBlessedModelResolver, ) from pipeline.components.HFPusher.component import HFPusher # from pipeline.components.HFPusher.component import HFSpaceConfig def create_pipeline( pipeline_name: Text, pipeline_root: Text, data_path: Text, schema_path: Text, modules: Dict[Text, Text], eval_configs: tfma.EvalConfig, metadata_connection_config: Optional[metadata_store_pb2.ConnectionConfig] = None, ai_platform_training_args: Optional[Dict[Text, Text]] = None, ai_platform_tuner_args: Optional[Dict[Text, Text]] = None, tuner_args: tuner_pb2.TuneArgs = None, ai_platform_serving_args: Optional[Dict[Text, Any]] = None, example_gen_beam_args: Optional[List] = None, transform_beam_args: Optional[List] = None, hf_pusher_args: Optional[Dict[Text, Any]] = None, ) -> tfx.dsl.Pipeline: components = [] input_config = example_gen_pb2.Input( splits=[ example_gen_pb2.Input.Split(name="train", pattern="train-*.tfrec"), example_gen_pb2.Input.Split(name="eval", pattern="val-*.tfrec"), ] ) example_gen = ImportExampleGen(input_base=data_path, input_config=input_config) if example_gen_beam_args is not None: example_gen.with_beam_pipeline_args(example_gen_beam_args) components.append(example_gen) statistics_gen = StatisticsGen(examples=example_gen.outputs["examples"]) components.append(statistics_gen) schema_gen = tfx.components.ImportSchemaGen(schema_file=schema_path) components.append(schema_gen) example_validator = ExampleValidator( statistics=statistics_gen.outputs["statistics"], schema=schema_gen.outputs["schema"], ) components.append(example_validator) transform_args = { "examples": example_gen.outputs["examples"], "schema": schema_gen.outputs["schema"], "preprocessing_fn": modules["preprocessing_fn"], } transform = Transform(**transform_args) if transform_beam_args is not None: transform.with_beam_pipeline_args(transform_beam_args) components.append(transform) tuner = VertexTuner( tuner_fn=modules["tuner_fn"], examples=transform.outputs["transformed_examples"], transform_graph=transform.outputs["transform_graph"], tune_args=tuner_args, custom_config=ai_platform_tuner_args, ) components.append(tuner) trainer_args = { "run_fn": modules["training_fn"], "transformed_examples": transform.outputs["transformed_examples"], "transform_graph": transform.outputs["transform_graph"], "schema": schema_gen.outputs["schema"], "hyperparameters": tuner.outputs["best_hyperparameters"], "custom_config": ai_platform_training_args, } trainer = VertexTrainer(**trainer_args) components.append(trainer) model_resolver = resolver.Resolver( strategy_class=LatestBlessedModelResolver, model=Channel(type=Model), model_blessing=Channel(type=ModelBlessing), ).with_id("latest_blessed_model_resolver") components.append(model_resolver) evaluator = Evaluator( examples=example_gen.outputs["examples"], model=trainer.outputs["model"], baseline_model=model_resolver.outputs["model"], eval_config=eval_configs, ) components.append(evaluator) pusher_args = { "model": trainer.outputs["model"], "model_blessing": evaluator.outputs["blessing"], "custom_config": ai_platform_serving_args, } pusher = VertexPusher(**pusher_args) # pylint: disable=unused-variable components.append(pusher) hf_pusher_args['model'] = trainer.outputs["model"] hf_pusher_args['model_blessing'] = evaluator.outputs["blessing"] hf_pusher = HFPusher(**hf_pusher_args) components.append(hf_pusher) return pipeline.Pipeline( pipeline_name=pipeline_name, pipeline_root=pipeline_root, components=components, enable_cache=True, metadata_connection_config=metadata_connection_config, )
deep-diver/mlops-hf-tf-vision-models
hf_integration/pipeline/local_pipeline.py
from typing import Dict, Optional, Text import tensorflow_model_analysis as tfma from tfx import v1 as tfx from ml_metadata.proto import metadata_store_pb2 from tfx.proto import example_gen_pb2 from tfx.components import ImportExampleGen from tfx.components import StatisticsGen from tfx.components import ExampleValidator from tfx.components import Transform from tfx.components import Tuner from tfx.components import Trainer from tfx.components import Evaluator from tfx.components import Pusher from tfx.orchestration import pipeline from tfx.proto import example_gen_pb2 from tfx.types import Channel from tfx.types.standard_artifacts import Model from tfx.types.standard_artifacts import ModelBlessing from tfx.dsl.components.common import resolver from tfx.dsl.experimental.latest_blessed_model_resolver import ( LatestBlessedModelResolver, ) from pipeline.components.HFPusher.component import HFPusher # from pipeline.components.HFPusher.component import HFSpaceConfig def create_pipeline( pipeline_name: Text, pipeline_root: Text, data_path: Text, schema_path: Text, modules: Dict[Text, Text], hyperparameters: Dict[Text, Text], eval_configs: tfma.EvalConfig, serving_model_dir: Text, metadata_connection_config: Optional[metadata_store_pb2.ConnectionConfig] = None, ) -> tfx.dsl.Pipeline: components = [] input_config = example_gen_pb2.Input( splits=[ example_gen_pb2.Input.Split(name="train", pattern="train-00-*.tfrec"), example_gen_pb2.Input.Split(name="eval", pattern="val-00-*.tfrec"), ] ) example_gen = ImportExampleGen(input_base=data_path, input_config=input_config) components.append(example_gen) statistics_gen = StatisticsGen(examples=example_gen.outputs["examples"]) components.append(statistics_gen) schema_gen = tfx.components.ImportSchemaGen(schema_file=schema_path) components.append(schema_gen) example_validator = ExampleValidator( statistics=statistics_gen.outputs["statistics"], schema=schema_gen.outputs["schema"], ) components.append(example_validator) transform_args = { "examples": example_gen.outputs["examples"], "schema": schema_gen.outputs["schema"], "preprocessing_fn": modules["preprocessing_fn"], } transform = Transform(**transform_args) components.append(transform) tuner = Tuner( tuner_fn=modules["tuner_fn"], examples=transform.outputs["transformed_examples"], schema=schema_gen.outputs["schema"], transform_graph=transform.outputs["transform_graph"], custom_config={"hyperparameters": hyperparameters}, ) components.append(tuner) trainer_args = { "run_fn": modules["training_fn"], "transformed_examples": transform.outputs["transformed_examples"], "transform_graph": transform.outputs["transform_graph"], "schema": schema_gen.outputs["schema"], "hyperparameters": tuner.outputs["best_hyperparameters"], "custom_config": {"is_local": True}, } trainer = Trainer(**trainer_args) components.append(trainer) model_resolver = resolver.Resolver( strategy_class=LatestBlessedModelResolver, model=Channel(type=Model), model_blessing=Channel(type=ModelBlessing), ).with_id("latest_blessed_model_resolver") components.append(model_resolver) evaluator = Evaluator( examples=example_gen.outputs["examples"], model=trainer.outputs["model"], baseline_model=model_resolver.outputs["model"], eval_config=eval_configs, ) components.append(evaluator) pusher_args = { "model": trainer.outputs["model"], "model_blessing": evaluator.outputs["blessing"], "push_destination": tfx.proto.PushDestination( filesystem=tfx.proto.PushDestination.Filesystem( base_directory=serving_model_dir ) ), } pusher = Pusher(**pusher_args) # pylint: disable=unused-variable components.append(pusher) # pusher_args = { # "model": trainer.outputs["model"], # "model_blessing": evaluator.outputs["blessing"], # "username": "chansung", # "access_token": "<REPLACE_WITH_YOUR_OWN_TOKEN>", # "repo_name": "vit-e2e-pipeline-hf-integration", # } # hf_pusher = HFPusher(**pusher_args) # components.append(hf_pusher) return pipeline.Pipeline( pipeline_name=pipeline_name, pipeline_root=pipeline_root, components=components, enable_cache=False, metadata_connection_config=metadata_connection_config, )
deep-diver/mlops-hf-tf-vision-models
intermediate/modules/ViT.py
import tensorflow as tf from transformers import TFViTForImageClassification from .common import LABELS from .common import PRETRAIN_CHECKPOINT from .utils import INFO def build_model(): id2label = {str(i): c for i, c in enumerate(LABELS)} label2id = {c: str(i) for i, c in enumerate(LABELS)} model = TFViTForImageClassification.from_pretrained( PRETRAIN_CHECKPOINT, num_labels=len(LABELS), label2id=label2id, id2label=id2label, ) model.layers[0].trainable = False loss = tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True) model.compile(optimizer="adam", loss=loss, metrics=["accuracy"]) INFO(model.summary()) return model
deep-diver/mlops-hf-tf-vision-models
intermediate/modules/common.py
IMAGE_TFREC_KEY = "image" IMAGE_SHAPE_TFREC_KEY = "image_shape" LABEL_TFREC_KEY = "label" MODEL_INPUT_IMAGE_KEY = "pixel_values" MODEL_INPUT_LABEL_KEY = "labels" IMAGE_MODEL_KEY = "pixel_values" LABEL_MODEL_KEY = "labels" CONCRETE_INPUT = "pixel_values" PRETRAIN_CHECKPOINT = "google/vit-base-patch16-224-in21k" LABELS = ["angular_leaf_spot", "bean_rust", "healthy"]
deep-diver/mlops-hf-tf-vision-models
intermediate/modules/hyperparams.py
EPOCHS = 1 BATCH_SIZE = 32 TRAIN_BATCH_SIZE = 32 EVAL_BATCH_SIZE = 32 TRAIN_LENGTH = 1034 EVAL_LENGTH = 128 INPUT_IMG_SIZE = 224
deep-diver/mlops-hf-tf-vision-models
intermediate/modules/preprocessing.py
import tensorflow as tf from .common import IMAGE_TFREC_KEY, LABEL_TFREC_KEY from .common import IMAGE_MODEL_KEY, LABEL_MODEL_KEY from .hyperparams import INPUT_IMG_SIZE def preprocessing_fn(inputs): """tf.transform's callback function for preprocessing inputs. Args: inputs: map from feature keys to raw not-yet-transformed features. Returns: Map from string feature key to transformed feature operations. """ # print(inputs) outputs = {} inputs[IMAGE_TFREC_KEY] = tf.image.resize( inputs[IMAGE_TFREC_KEY], [INPUT_IMG_SIZE, INPUT_IMG_SIZE] ) inputs[IMAGE_TFREC_KEY] = inputs[IMAGE_TFREC_KEY] / 255.0 inputs[IMAGE_TFREC_KEY] = tf.transpose(inputs[IMAGE_TFREC_KEY], [0, 3, 1, 2]) outputs[IMAGE_MODEL_KEY] = inputs[IMAGE_TFREC_KEY] outputs[LABEL_MODEL_KEY] = inputs[LABEL_TFREC_KEY] return outputs
deep-diver/mlops-hf-tf-vision-models
intermediate/modules/signatures.py
import tensorflow as tf from transformers import ViTFeatureExtractor from .common import PRETRAIN_CHECKPOINT from .common import CONCRETE_INPUT feature_extractor = ViTFeatureExtractor.from_pretrained(PRETRAIN_CHECKPOINT) def _normalize_img( img, mean=feature_extractor.image_mean, std=feature_extractor.image_std ): img = img / 255 mean = tf.constant(mean) std = tf.constant(std) return (img - mean) / std def _preprocess_serving(string_input): decoded_input = tf.io.decode_base64(string_input) decoded = tf.io.decode_jpeg(decoded_input, channels=3) resized = tf.image.resize(decoded, size=(224, 224)) normalized = _normalize_img(resized) normalized = tf.transpose( normalized, (2, 0, 1) ) # Since HF models are channel-first. return normalized @tf.function(input_signature=[tf.TensorSpec([None], tf.string)]) def _preprocess_fn(string_input): decoded_images = tf.map_fn( _preprocess_serving, string_input, dtype=tf.float32, back_prop=False ) return {CONCRETE_INPUT: decoded_images} def model_exporter(model: tf.keras.Model): m_call = tf.function(model.call).get_concrete_function( tf.TensorSpec(shape=[None, 3, 224, 224], dtype=tf.float32, name=CONCRETE_INPUT) ) @tf.function(input_signature=[tf.TensorSpec([None], tf.string)]) def serving_fn(string_input): labels = tf.constant(list(model.config.id2label.values()), dtype=tf.string) images = _preprocess_fn(string_input) predictions = m_call(**images) indices = tf.argmax(predictions.logits, axis=1) pred_source = tf.gather(params=labels, indices=indices) probs = tf.nn.softmax(predictions.logits, axis=1) pred_confidence = tf.reduce_max(probs, axis=1) return {"label": pred_source, "confidence": pred_confidence} return serving_fn
deep-diver/mlops-hf-tf-vision-models
intermediate/modules/train.py
import tensorflow_transform as tft from tfx.components.trainer.fn_args_utils import FnArgs from .train_data import input_fn from .ViT import build_model from .signatures import model_exporter from .hyperparams import TRAIN_BATCH_SIZE, EVAL_BATCH_SIZE from .hyperparams import TRAIN_LENGTH, EVAL_LENGTH from .hyperparams import EPOCHS def run_fn(fn_args: FnArgs): tf_transform_output = tft.TFTransformOutput(fn_args.transform_output) train_dataset = input_fn( fn_args.train_files, fn_args.data_accessor, tf_transform_output, is_train=True, batch_size=TRAIN_BATCH_SIZE, ) eval_dataset = input_fn( fn_args.eval_files, fn_args.data_accessor, tf_transform_output, is_train=False, batch_size=EVAL_BATCH_SIZE, ) model = build_model() model.fit( train_dataset, steps_per_epoch=TRAIN_LENGTH // TRAIN_BATCH_SIZE, validation_data=eval_dataset, validation_steps=EVAL_LENGTH // TRAIN_BATCH_SIZE, epochs=EPOCHS, ) model.save( fn_args.serving_model_dir, save_format="tf", signatures=model_exporter(model) )
deep-diver/mlops-hf-tf-vision-models
intermediate/modules/train_data.py
from typing import List import tensorflow as tf import tensorflow_transform as tft from tfx_bsl.tfxio import dataset_options from tfx.components.trainer.fn_args_utils import DataAccessor from .utils import INFO from .common import LABEL_MODEL_KEY from .hyperparams import BATCH_SIZE def input_fn( file_pattern: List[str], data_accessor: DataAccessor, tf_transform_output: tft.TFTransformOutput, is_train: bool = False, batch_size: int = BATCH_SIZE, ) -> tf.data.Dataset: INFO(f"Reading data from: {file_pattern}") dataset = data_accessor.tf_dataset_factory( file_pattern, dataset_options.TensorFlowDatasetOptions( batch_size=batch_size, label_key=LABEL_MODEL_KEY ), tf_transform_output.transformed_metadata.schema, ) return dataset
deep-diver/mlops-hf-tf-vision-models
intermediate/modules/utils.py
import absl def INFO(text: str): absl.logging.info(text)
deep-diver/mlops-hf-tf-vision-models
intermediate/pipeline/configs.py
import os # pylint: disable=unused-import import tfx.extensions.google_cloud_ai_platform.constants as vertex_const import tfx.extensions.google_cloud_ai_platform.trainer.executor as vertex_training_const PIPELINE_NAME = "vit-e2e-pipeline-intermediate" try: import google.auth # pylint: disable=g-import-not-at-top # pytype: disable=import-error try: _, GOOGLE_CLOUD_PROJECT = google.auth.default() except google.auth.exceptions.DefaultCredentialsError: GOOGLE_CLOUD_PROJECT = "gcp-ml-172005" except ImportError: GOOGLE_CLOUD_PROJECT = "gcp-ml-172005" GOOGLE_CLOUD_REGION = "us-central1" GCS_BUCKET_NAME = GOOGLE_CLOUD_PROJECT + "-complete-mlops" PIPELINE_IMAGE = f"gcr.io/{GOOGLE_CLOUD_PROJECT}/{PIPELINE_NAME}" OUTPUT_DIR = os.path.join("gs://", GCS_BUCKET_NAME) PIPELINE_ROOT = os.path.join(OUTPUT_DIR, "tfx_pipeline_output", PIPELINE_NAME) DATA_PATH = "gs://beans-lowres/tfrecords/" SCHEMA_PATH = "pipeline/schema.pbtxt" TRAINING_FN = "modules.train.run_fn" PREPROCESSING_FN = "modules.preprocessing.preprocessing_fn" EXAMPLE_GEN_BEAM_ARGS = None TRANSFORM_BEAM_ARGS = None GCP_AI_PLATFORM_TRAINING_ARGS = { vertex_const.ENABLE_VERTEX_KEY: True, vertex_const.VERTEX_REGION_KEY: GOOGLE_CLOUD_REGION, vertex_training_const.TRAINING_ARGS_KEY: { "project": GOOGLE_CLOUD_PROJECT, "worker_pool_specs": [ { "machine_spec": { "machine_type": "n1-standard-4", "accelerator_type": "NVIDIA_TESLA_K80", "accelerator_count": 1, }, "replica_count": 1, "container_spec": { "image_uri": PIPELINE_IMAGE, }, } ], }, "use_gpu": True, } fullres_data = os.environ.get("FULL_RES_DATA", "false") if fullres_data.lower() == "true": DATA_PATH = "gs://beans-fullres/tfrecords/" DATAFLOW_SERVICE_ACCOUNT = "csp-gde-dataflow@gcp-ml-172005.iam.gserviceaccount.com" DATAFLOW_MACHINE_TYPE = "n1-standard-4" DATAFLOW_MAX_WORKERS = 4 DATAFLOW_DISK_SIZE_GB = 100 EXAMPLE_GEN_BEAM_ARGS = [ "--runner=DataflowRunner", "--project=" + GOOGLE_CLOUD_PROJECT, "--region=" + GOOGLE_CLOUD_REGION, "--service_account_email=" + DATAFLOW_SERVICE_ACCOUNT, "--machine_type=" + DATAFLOW_MACHINE_TYPE, "--experiments=use_runner_v2", "--max_num_workers=" + str(DATAFLOW_MAX_WORKERS), "--disk_size_gb=" + str(DATAFLOW_DISK_SIZE_GB), ] TRANSFORM_BEAM_ARGS = [ "--runner=DataflowRunner", "--project=" + GOOGLE_CLOUD_PROJECT, "--region=" + GOOGLE_CLOUD_REGION, "--service_account_email=" + DATAFLOW_SERVICE_ACCOUNT, "--machine_type=" + DATAFLOW_MACHINE_TYPE, "--experiments=use_runner_v2", "--max_num_workers=" + str(DATAFLOW_MAX_WORKERS), "--disk_size_gb=" + str(DATAFLOW_DISK_SIZE_GB), "--worker_harness_container_image=" + PIPELINE_IMAGE, ] GCP_AI_PLATFORM_TRAINING_ARGS[vertex_training_const.TRAINING_ARGS_KEY][ "worker_pool_specs" ] = [ { "machine_spec": { "machine_type": "n1-standard-8", "accelerator_type": "NVIDIA_TESLA_V100", "accelerator_count": 1, }, "replica_count": 1, "container_spec": { "image_uri": PIPELINE_IMAGE, }, } ] GCP_AI_PLATFORM_SERVING_ARGS = { vertex_const.ENABLE_VERTEX_KEY: True, vertex_const.VERTEX_REGION_KEY: GOOGLE_CLOUD_REGION, vertex_const.VERTEX_CONTAINER_IMAGE_URI_KEY: "us-docker.pkg.dev/vertex-ai/prediction/tf2-cpu.2-8:latest", vertex_const.SERVING_ARGS_KEY: { "project_id": GOOGLE_CLOUD_PROJECT, "deployed_model_display_name": PIPELINE_NAME.replace("-", "_"), "endpoint_name": "prediction-" + PIPELINE_NAME.replace("-", "_"), "traffic_split": {"0": 100}, "machine_type": "n1-standard-4", "min_replica_count": 1, "max_replica_count": 1, }, }
deep-diver/mlops-hf-tf-vision-models
intermediate/pipeline/kubeflow_pipeline.py
from typing import Any, Dict, List, Optional, Text from tfx import v1 as tfx from ml_metadata.proto import metadata_store_pb2 from tfx.proto import example_gen_pb2 from tfx.components import ImportExampleGen from tfx.components import StatisticsGen from tfx.components import ExampleValidator from tfx.components import Transform from tfx.extensions.google_cloud_ai_platform.trainer.component import ( Trainer as VertexTrainer, ) from tfx.extensions.google_cloud_ai_platform.pusher.component import ( Pusher as VertexPusher, ) from tfx.orchestration import pipeline from tfx.proto import example_gen_pb2 def create_pipeline( pipeline_name: Text, pipeline_root: Text, data_path: Text, schema_path: Text, modules: Dict[Text, Text], metadata_connection_config: Optional[metadata_store_pb2.ConnectionConfig] = None, ai_platform_training_args: Optional[Dict[Text, Text]] = None, ai_platform_serving_args: Optional[Dict[Text, Any]] = None, example_gen_beam_args: Optional[List] = None, transform_beam_args: Optional[List] = None, ) -> tfx.dsl.Pipeline: components = [] input_config = example_gen_pb2.Input( splits=[ example_gen_pb2.Input.Split(name="train", pattern="train-*.tfrec"), example_gen_pb2.Input.Split(name="eval", pattern="val-*.tfrec"), ] ) example_gen = ImportExampleGen(input_base=data_path, input_config=input_config) if example_gen_beam_args is not None: example_gen.with_beam_pipeline_args(example_gen_beam_args) components.append(example_gen) statistics_gen = StatisticsGen(examples=example_gen.outputs["examples"]) components.append(statistics_gen) schema_gen = tfx.components.ImportSchemaGen(schema_file=schema_path) components.append(schema_gen) example_validator = ExampleValidator( statistics=statistics_gen.outputs["statistics"], schema=schema_gen.outputs["schema"], ) components.append(example_validator) transform_args = { "examples": example_gen.outputs["examples"], "schema": schema_gen.outputs["schema"], "preprocessing_fn": modules["preprocessing_fn"], } transform = Transform(**transform_args) if transform_beam_args is not None: transform.with_beam_pipeline_args(transform_beam_args) components.append(transform) trainer_args = { "run_fn": modules["training_fn"], "transformed_examples": transform.outputs["transformed_examples"], "transform_graph": transform.outputs["transform_graph"], "schema": schema_gen.outputs["schema"], "custom_config": ai_platform_training_args, } trainer = VertexTrainer(**trainer_args) components.append(trainer) pusher_args = { "model": trainer.outputs["model"], "custom_config": ai_platform_serving_args, } pusher = VertexPusher(**pusher_args) # pylint: disable=unused-variable components.append(pusher) return pipeline.Pipeline( pipeline_name=pipeline_name, pipeline_root=pipeline_root, components=components, enable_cache=True, metadata_connection_config=metadata_connection_config, )
deep-diver/mlops-hf-tf-vision-models
intermediate/pipeline/local_pipeline.py
from typing import Dict, Optional, Text from tfx import v1 as tfx from ml_metadata.proto import metadata_store_pb2 from tfx.proto import example_gen_pb2 from tfx.components import ImportExampleGen from tfx.components import StatisticsGen from tfx.components import ExampleValidator from tfx.components import Transform from tfx.components import Trainer from tfx.components import Pusher from tfx.orchestration import pipeline from tfx.proto import example_gen_pb2 def create_pipeline( pipeline_name: Text, pipeline_root: Text, data_path: Text, schema_path: Text, modules: Dict[Text, Text], serving_model_dir: Text, metadata_connection_config: Optional[metadata_store_pb2.ConnectionConfig] = None, ) -> tfx.dsl.Pipeline: components = [] input_config = example_gen_pb2.Input( splits=[ example_gen_pb2.Input.Split(name="train", pattern="train-00-*.tfrec"), example_gen_pb2.Input.Split(name="eval", pattern="val-00-*.tfrec"), ] ) example_gen = ImportExampleGen(input_base=data_path, input_config=input_config) components.append(example_gen) statistics_gen = StatisticsGen(examples=example_gen.outputs["examples"]) components.append(statistics_gen) schema_gen = tfx.components.ImportSchemaGen(schema_file=schema_path) components.append(schema_gen) example_validator = ExampleValidator( statistics=statistics_gen.outputs["statistics"], schema=schema_gen.outputs["schema"], ) components.append(example_validator) transform_args = { "examples": example_gen.outputs["examples"], "schema": schema_gen.outputs["schema"], "preprocessing_fn": modules["preprocessing_fn"], } transform = Transform(**transform_args) components.append(transform) trainer_args = { "run_fn": modules["training_fn"], "transformed_examples": transform.outputs["transformed_examples"], "transform_graph": transform.outputs["transform_graph"], "schema": schema_gen.outputs["schema"], } trainer = Trainer(**trainer_args) components.append(trainer) pusher_args = { "model": trainer.outputs["model"], "push_destination": tfx.proto.PushDestination( filesystem=tfx.proto.PushDestination.Filesystem( base_directory=serving_model_dir ) ), } pusher = Pusher(**pusher_args) # pylint: disable=unused-variable components.append(pusher) return pipeline.Pipeline( pipeline_name=pipeline_name, pipeline_root=pipeline_root, components=components, enable_cache=False, metadata_connection_config=metadata_connection_config, )
deep-diver/mlops-hf-tf-vision-models
hf_integration/app/gradio/app.py
import gradio as gr import numpy as np from PIL import Image import tensorflow as tf from transformers import ViTFeatureExtractor from huggingface_hub import from_pretrained_keras PRETRAIN_CHECKPOINT = "google/vit-base-patch16-224-in21k" feature_extractor = ViTFeatureExtractor.from_pretrained(PRETRAIN_CHECKPOINT) # $MODEL_REPO_ID should be like chansung/test-vit # $MODEL_VERSION MODEL_CKPT = "$MODEL_REPO_ID@$MODEL_VERSION" MODEL = from_pretrained_keras(MODEL_CKPT) RESOLTUION = 224 labels = [] with open(r"labels.txt", "r") as fp: for line in fp: labels.append(line[:-1]) def normalize_img( img, mean=feature_extractor.image_mean, std=feature_extractor.image_std ): img = img / 255 mean = tf.constant(mean) std = tf.constant(std) return (img - mean) / std def preprocess_input(image: Image) -> tf.Tensor: image = np.array(image) image = tf.convert_to_tensor(image) image = tf.image.resize(image, (RESOLTUION, RESOLTUION)) image = normalize_img(image) image = tf.transpose( image, (2, 0, 1) ) # Since HF models are channel-first. return { "pixel_values": tf.expand_dims(image, 0) } def get_predictions(image: Image) -> tf.Tensor: preprocessed_image = preprocess_input(image) prediction = MODEL.predict(preprocessed_image) probs = tf.nn.softmax(prediction['logits'], axis=1) confidences = {labels[i]: float(probs[0][i]) for i in range(3)} return confidences title = "Simple demo for a Image Classification of the Beans Dataset with HF ViT model" demo = gr.Interface( get_predictions, gr.inputs.Image(type="pil"), gr.outputs.Label(num_top_classes=3), allow_flagging="never", title=title, ) demo.launch(debug=True)
deep-diver/mlops-hf-tf-vision-models
hf_integration/pipeline/components/HFPusher/__init__.py
# Copyright 2022 The TensorFlow Authors. All Rights Reserved. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. # ==============================================================================
deep-diver/mlops-hf-tf-vision-models
hf_integration/pipeline/components/HFPusher/component.py
# Copyright 2022 The TensorFlow Authors. All Rights Reserved. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. # ============================================================================== """HuggingFace(HF) Pusher TFX Component. The HFPusher is used to push model and prototype application to HuggingFace Hub. """ from typing import Text, Dict, Any, Optional from tfx import types from tfx.dsl.components.base import base_component, executor_spec from tfx.types import standard_artifacts from tfx.types.component_spec import ChannelParameter, ExecutionParameter from pipeline.components.HFPusher import executor MODEL_KEY = "model" PUSHED_MODEL_KEY = "pushed_model" MODEL_BLESSING_KEY = "model_blessing" class HFPusherSpec(types.ComponentSpec): """ComponentSpec for TFX HFPusher Component.""" PARAMETERS = { "username": ExecutionParameter(type=str), "access_token": ExecutionParameter(type=str), "repo_name": ExecutionParameter(type=str), "space_config": ExecutionParameter(type=Dict[Text, Any], optional=True), } INPUTS = { MODEL_KEY: ChannelParameter(type=standard_artifacts.Model, optional=True), MODEL_BLESSING_KEY: ChannelParameter( type=standard_artifacts.ModelBlessing, optional=True ), } OUTPUTS = { PUSHED_MODEL_KEY: ChannelParameter(type=standard_artifacts.PushedModel), } class HFPusher(base_component.BaseComponent): """Component for pushing model and application to HuggingFace Hub. The `HFPusher` is a [TFX Component](https://www.tensorflow.org/tfx /guide/understanding_tfx_pipelines#component), and its primary pur pose is to push a model from an upstream component such as [`Train er`](https://www.tensorflow.org/tfx/guide/trainer) to HuggingFace Model Hub. It also provides a secondary feature that pushes an app lication to HuggingFace Space Hub. """ SPEC_CLASS = HFPusherSpec EXECUTOR_SPEC = executor_spec.ExecutorClassSpec(executor.Executor) def __init__( self, username: str, access_token: str, repo_name: str, space_config: Optional[Dict[Text, Any]] = None, model: Optional[types.Channel] = None, model_blessing: Optional[types.Channel] = None, ): """The HFPusher TFX component. HFPusher pushes a trained or blessed model to HuggingFace Model Hub. This is designed to work as a downstream component of Trainer and o ptionally Evaluator(optional) components. Trainer gives trained mod el, and Evaluator gives information whether the trained model is bl essed or not after evaluation of the model. HFPusher component only publishes a model when it is blessed. If Evaluator is not specified, the input model will always be pushed. Args: username: the ID of HuggingFace Hub access_token: the access token obtained from HuggingFace Hub for the given username. Refer to [this document](https://huggingface.co/ docs/hub/security-tokens) to know how to obtain one. repo_name: the name of Model Hub repository where the model will be pushed. This should be unique name under the username within th e Model Hub. repository is identified as {username}/{repo_name}. space_config: optional configurations set when to push an application to HuggingFace Space Hub. This is a dictionary, and the following information could be set. app_path: the path where the application related files are stored. this should follow the form either of app.gradio.segmentation or app/gradio/segmentation. This is a required parameter when space_config is set. This could be a local or GCS paths. space_sdk: Space Hub supports gradio, streamit, and static types of application. The default is set to gradio. placeholders: placeholders to replace in every files under the a pp_path. This is used to replace special string with the mod el related values. If this is not set, the default placehold ers will be used as follows. ``` placeholders = { "MODEL_REPO_ID" : "$MODEL_REPO_ID", "MODEL_REPO_URL": "$MODEL_REPO_URL", "MODEL_VERSION" : "$MODEL_VERSION", } ``` In this case, "$MODEL_REPO_ID", "$MODEL_REPO_URL", "$MODEL_VE RSION" strings will be replaced with appropriate values at ru ntime. If placeholders are set, custom strings will be used. repo_name: the name of Space Hub repository where the application will be pushed. This should be unique name under the username within the Space Hub. repository is identified as {username}/ {repo_name}. If this is not set, the same name to the Model H ub repository will be used. model: a TFX input channel containing a Model artifact. this is usually comes from the standard [`Trainer`] (https://www.tensorflow.org/tfx/guide/trainer) component. model_blessing: a TFX input channel containing a ModelBlessing artifact. this is usually comes from the standard [`Evaluator`] (https://www.tensorflow.org/tfx/guide/evaluator) component. Returns: a TFX output channel containing a PushedModel artifact. It contains information where the model is published at and whether the model is pushed or not. Raises: RuntimeError: if app_path is not set when space_config is provided. Example: Basic usage example: ```py trainer = Trainer(...) evaluator = Evaluator(...) hf_pusher = HFPusher( username="chansung", access_token=<YOUR-HUGGINGFACE-ACCESS-TOKEN>, repo_name="my-model", model=trainer.outputs["model"], model_blessing=evaluator.outputs["blessing"], space_config={ "app_path": "apps.gradio.semantic_segmentation" } ) ``` """ pushed_model = types.Channel(type=standard_artifacts.PushedModel) spec = HFPusherSpec( username=username, access_token=access_token, repo_name=repo_name, space_config=space_config, model=model, model_blessing=model_blessing, pushed_model=pushed_model, ) super().__init__(spec=spec)
deep-diver/mlops-hf-tf-vision-models
hf_integration/pipeline/components/HFPusher/component_test.py
# Copyright 2022 The TensorFlow Authors. All Rights Reserved. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. # ============================================================================== """Tests for TFX HuggingFace Pusher Custom Component.""" import tensorflow as tf from tfx.types import standard_artifacts from tfx.types import channel_utils from pipeline.components.HFPusher.component import HFPusher class HFPusherTest(tf.test.TestCase): def testConstruct(self): test_model = channel_utils.as_channel([standard_artifacts.Model()]) hf_pusher = HFPusher( username="test_username", access_token="test_access_token", repo_name="test_repo_name", model=test_model, ) self.assertEqual( standard_artifacts.PushedModel.TYPE_NAME, hf_pusher.outputs["pushed_model"].type_name, ) if __name__ == "__main__": tf.test.main()
deep-diver/mlops-hf-tf-vision-models
hf_integration/pipeline/components/HFPusher/executor.py
# Copyright 2022 The TensorFlow Authors. All Rights Reserved. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. # ============================================================================== """HF Pusher TFX Component Executor. The HF Pusher Executor calls the workflow handler runner.deploy_model_for_hf_hub(). """ import ast import time from typing import Any, Dict, List from tfx import types from tfx.components.pusher import executor as tfx_pusher_executor from tfx.types import artifact_utils, standard_component_specs from pipeline.components.HFPusher import runner _USERNAME_KEY = "username" _ACCESS_TOKEN_KEY = "access_token" _REPO_NAME_KEY = "repo_name" _SPACE_CONFIG_KEY = "space_config" class Executor(tfx_pusher_executor.Executor): """Pushes a model and an app to HuggingFace Model and Space Hubs respectively""" def Do( self, input_dict: Dict[str, List[types.Artifact]], output_dict: Dict[str, List[types.Artifact]], exec_properties: Dict[str, Any], ): """Overrides the tfx_pusher_executor to leverage some of utility methods Args: input_dict: Input dict from input key to a list of artifacts, including: - model_export: a TFX input channel containing a Model artifact. - model_blessing: a TFX input channel containing a ModelBlessing artifact. output_dict: Output dict from key to a list of artifacts, including: - pushed_model: a TFX output channel containing a PushedModel arti fact. It contains information where the model is published at an d whether the model is pushed or not. furthermore, pushed model carries the following information. - pushed : integer value to denote if the model is pushed or not. This is set to 0 when the input model is not blessed, and it is set to 1 when the model is successfully pushed. - pushed_version : string value to indicate the current model ver sion. This is decided by time.time() Python built-in function. - repo_id : model repository ID where the model is pushed to. This follows the format of f"{username}/{repo_name}". - branch : branch name where the model is pushed to. The branch na me is automatically assigned to the same value of pushed_version. - commit_id : the id from the commit history (branch name could be sufficient to retreive a certain version of the model) of the mo del repository. - repo_url : model repository URL. It is something like f"https:// huggingface.co/{repo_id}/{branch}" - space_url : space repository URL. It is something like f"https:// huggingface.co/{repo_id}"f exec_properties: An optional dict of execution properties, including: - username: username of the HuggingFace user (can be an individual user or an organization) - access_token: access token value issued by HuggingFace for the s pecified username. - repo_name: the repository name to push the current version of the model to. The default value is same as the TFX pipeline name. - space_config: space_config carries additional values such as: - app_path : path where the application templates are in the cont ainer that runs the TFX pipeline. This is expressed either apps. gradio.img_classifier or apps/gradio.img_classifier. - repo_name : the repository name to push the application to. The default value is same as the TFX pipeline name - space_sdk : either gradio or streamlit. this will decide which a pplication framework to be used for the Space repository. The de fault value is gradio - placeholders : dictionary which placeholders to replace with mod el specific information. The keys represents describtions, and t he values represents the actual placeholders to replace in the f iles under the app_path. There are currently two predefined keys, and if placeholders is set to None, the default values will be used. """ self._log_startup(input_dict, output_dict, exec_properties) model_push = artifact_utils.get_single_instance( output_dict[standard_component_specs.PUSHED_MODEL_KEY] ) # if the model is not blessed if not self.CheckBlessing(input_dict): self._MarkNotPushed(model_push) return model_path = self.GetModelPath(input_dict) model_version_name = f"v{int(time.time())}" space_config = exec_properties.get(_SPACE_CONFIG_KEY, None) if space_config is not None: space_config = ast.literal_eval(space_config) pushed_properties = runner.deploy_model_for_hf_hub( username=exec_properties.get(_USERNAME_KEY, None), access_token=exec_properties.get(_ACCESS_TOKEN_KEY, None), repo_name=exec_properties.get(_REPO_NAME_KEY, None), space_config=space_config, model_path=model_path, model_version=model_version_name, ) self._MarkPushed(model_push, pushed_destination=pushed_properties["repo_url"]) for key in pushed_properties: value = pushed_properties[key] if key != "repo_url": model_push.set_string_custom_property(key, value)
deep-diver/mlops-hf-tf-vision-models
hf_integration/pipeline/components/HFPusher/runner.py
# Copyright 2022 The TensorFlow Authors. All Rights Reserved. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. # ============================================================================== """HuggingFace Pusher runner module. This module handles the workflow to publish machine learning model to HuggingFace Hub. """ from typing import Text, Any, Dict, Optional import tempfile import tensorflow as tf from absl import logging from tfx.utils import io_utils from huggingface_hub import Repository from huggingface_hub import HfApi from requests.exceptions import HTTPError _MODEL_REPO_KEY = "MODEL_REPO_ID" _MODEL_URL_KEY = "MODEL_REPO_URL" _MODEL_VERSION_KEY = "MODEL_VERSION" _DEFAULT_MODEL_REPO_PLACEHOLDER_KEY = "$MODEL_REPO_ID" _DEFAULT_MODEL_URL_PLACEHOLDER_KEY = "$MODEL_REPO_URL" _DEFAULT_MODEL_VERSION_PLACEHOLDER_KEY = "$MODEL_VERSION" def _replace_placeholders_in_files( root_dir: str, placeholder_to_replace: Dict[str, str] ): """Recursively open every files under the root_dir, and then replace special tokens with the given values in placeholder_ to_replace""" files = tf.io.gfile.listdir(root_dir) for file in files: path = tf.io.gfile.join(root_dir, file) if tf.io.gfile.isdir(path): _replace_placeholders_in_files(path, placeholder_to_replace) else: _replace_placeholders_in_file(path, placeholder_to_replace) def _replace_placeholders_in_file( filepath: str, placeholder_to_replace: Dict[str, str] ): """replace special tokens with the given values in placeholder_ to_replace. This function gets called by _replace_placeholders _in_files function""" with tf.io.gfile.GFile(filepath, "r") as f: source_code = f.read() for placeholder in placeholder_to_replace: source_code = source_code.replace( placeholder, placeholder_to_replace[placeholder] ) with tf.io.gfile.GFile(filepath, "w") as f: f.write(source_code) def _replace_placeholders( target_dir: str, placeholders: Dict[str, str], model_repo_id: str, model_repo_url: str, model_version: str, ): """set placeholder_to_replace before calling _replace_placeholde rs_in_files function""" if placeholders is None: placeholders = { _MODEL_REPO_KEY: _DEFAULT_MODEL_REPO_PLACEHOLDER_KEY, _MODEL_URL_KEY: _DEFAULT_MODEL_URL_PLACEHOLDER_KEY, _MODEL_VERSION_KEY: _DEFAULT_MODEL_VERSION_PLACEHOLDER_KEY, } placeholder_to_replace = { placeholders[_MODEL_REPO_KEY]: model_repo_id, placeholders[_MODEL_URL_KEY]: model_repo_url, placeholders[_MODEL_VERSION_KEY]: model_version, } _replace_placeholders_in_files(target_dir, placeholder_to_replace) def _replace_files(src_path, dst_path): """replace the contents(files/folders) of the repository with the latest contents""" not_to_delete = [".gitattributes", ".git"] inside_root_dst_path = tf.io.gfile.listdir(dst_path) for content_name in inside_root_dst_path: content = f"{dst_path}/{content_name}" if content_name not in not_to_delete: if tf.io.gfile.isdir(content): tf.io.gfile.rmtree(content) else: tf.io.gfile.remove(content) inside_root_src_path = tf.io.gfile.listdir(src_path) for content_name in inside_root_src_path: content = f"{src_path}/{content_name}" dst_content = f"{dst_path}/{content_name}" if tf.io.gfile.isdir(content): io_utils.copy_dir(content, dst_content) else: tf.io.gfile.copy(content, dst_content) def _create_remote_repo( access_token: str, repo_id: str, repo_type: str = "model", space_sdk: str = None ): """create a remote repository on HuggingFace Hub platform. HTTPError exception is raised when the repository already exists""" logging.info(f"repo_id: {repo_id}") try: HfApi().create_repo( token=access_token, repo_id=repo_id, repo_type=repo_type, space_sdk=space_sdk, ) except HTTPError: logging.warning( f"this warning is expected if {repo_id} repository already exists" ) def _clone_and_checkout( repo_url: str, local_path: str, access_token: str, version: Optional[str] = None ) -> Repository: """clone the remote repository to the given local_path""" repository = Repository( local_dir=local_path, clone_from=repo_url, use_auth_token=access_token ) if version is not None: repository.git_checkout(revision=version, create_branch_ok=True) return repository def _push_to_remote_repo(repo: Repository, commit_msg: str, branch: str = "main"): """push any changes to the remote repository""" repo.git_add(pattern=".", auto_lfs_track=True) repo.git_commit(commit_message=commit_msg) repo.git_push(upstream=f"origin {branch}") def deploy_model_for_hf_hub( username: str, access_token: str, repo_name: str, model_path: str, model_version: str, space_config: Optional[Dict[Text, Any]] = None, ) -> Dict[str, str]: """Executes ML model deployment workflow to HuggingFace Hub. Refer to the HFPusher component in component.py for generic description of each parame ter. This docstring only explains how the workflow works. step 1. push model to the Model Hub step 1-1. create a repository on the HuggingFace Hub. if there is an existing r epository with the given repo_name, that rpository will be overwritten. step 1-2. clone the created or existing remote repository to the local path. Al so, create a branch named with model version. step 1-3. remove every files under the cloned repository(local), and copies the model related files to the cloned local repository path. step 1-4. push the updated repository to the given branch of remote Model Hub. step 2. push application to the Space Hub step 2-1. create a repository on the HuggingFace Hub. if there is an existing r epository with the given repo_name, that rpository will be overwritten. step 2-2. copies directory where the application related files are stored to a temporary directory. Since the files could be hosted in GCS bucket, t his process ensures every necessary files are located in the local fil e system. step 2-3. replacek speical tokens in every files under the given directory. step 2-4. clone the created or existing remote repository to the local path. step 2-5. remove every files under the cloned repository(local), and copies the application related files to the cloned local repository path. step 2-6. push the updated repository to the remote Space Hub. note that the br anch is always set to "main", so that HuggingFace Space could build t he application automatically when pushed. """ outputs = {} # step 1 repo_url_prefix = "https://huggingface.co" repo_id = f"{username}/{repo_name}" repo_url = f"{repo_url_prefix}/{repo_id}" # step 1-1 _create_remote_repo(access_token=access_token, repo_id=repo_id) logging.info(f"remote repository at {repo_url} is prepared") # step 1-2 local_path = "hf_model" repository = _clone_and_checkout( repo_url=repo_url, local_path=local_path, access_token=access_token, version=model_version, ) logging.info( f"remote repository is cloned, and new branch {model_version} is created" ) # step 1-3 _replace_files(model_path, local_path) logging.info( "current version of the model is copied to the cloned local repository" ) # step 1-4 _push_to_remote_repo( repo=repository, commit_msg=f"updload new version({model_version})", branch=model_version, ) logging.info("updates are pushed to the remote repository") outputs["repo_id"] = repo_id outputs["branch"] = model_version outputs["commit_id"] = f"{repository.git_head_hash()}" outputs["repo_url"] = repo_url # step 2 if space_config is not None: if "app_path" not in space_config: raise RuntimeError( f"the app_path is not provided. " f"app_path is required when space_config is set." ) model_repo_id = repo_id model_repo_url = repo_url if "repo_name" in space_config: repo_id = f"{username}/{repo_name}" repo_url = f"{repo_url_prefix}/{repo_id}" else: repo_url = f"{repo_url_prefix}/spaces/{repo_id}" app_path = space_config["app_path"] app_path = app_path.replace(".", "/") # step 2-1 _create_remote_repo( access_token=access_token, repo_id=repo_id, repo_type="space", space_sdk=space_config["space_sdk"] if "space_sdk" in space_config else "gradio", ) # step 2-2 tmp_dir = tempfile.gettempdir() io_utils.copy_dir(app_path, tmp_dir) # step 2-3 _replace_placeholders( target_dir=tmp_dir, placeholders=space_config["placeholders"] if "placeholders" in space_config else None, model_repo_id=model_repo_id, model_repo_url=model_repo_url, model_version=model_version, ) # step 2-4 local_path = "hf_space" repository = _clone_and_checkout( repo_url=repo_url, local_path=local_path, access_token=access_token, ) # step 2-5 _replace_files(tmp_dir, local_path) # step 2-6 _push_to_remote_repo( repo=repository, commit_msg=f"upload {model_version} model", ) outputs["space_url"] = repo_url return outputs
deep-diver/semantic-segmentation-ml-pipeline
notebooks/gradio_demo_pets.ipynb
from huggingface_hub import from_pretrained_keras model_ckpt = "chansung/segmentation-training-pipeline@v1667662600" MODEL = from_pretrained_keras(model_ckpt)from PIL import Image import numpy as np import tensorflow as tf RESOLTUION = 128 # MODEL = tf.keras.models.load_model(model_path) def preprocess_input(image: Image) -> tf.Tensor: image = np.array(image) image = tf.convert_to_tensor(image) image = tf.image.resize(image, (RESOLTUION, RESOLTUION)) image = image / 255 return tf.expand_dims(image, 0) # The below utilities (sidewalk_palette(), get_seg_overlay()) are from: # https://github.com/deep-diver/semantic-segmentation-ml-pipeline/blob/main/notebooks/inference_from_SavedModel.ipynb def pets_palette(): """Pets palette that maps each class to RGB values.""" return [ [ 0, 10, 146], [ 38, 0, 44], [255, 232, 0], ] def get_seg_overlay(image, seg): color_seg = np.zeros( (seg.shape[0], seg.shape[1], 3), dtype=np.uint8 ) # height, width, 3 palette = np.array(pets_palette()) for label, color in enumerate(palette): color_seg[seg == label, :] = color # Show image + mask img = np.array(image) * 0.5 + color_seg * 0.5 img *= 255 img = np.clip(img, 0, 255) img = img.astype(np.uint8) return img def run_model(image: Image) -> tf.Tensor: preprocessed_image = preprocess_input(image) prediction = MODEL.predict(preprocessed_image) seg_mask = tf.math.argmax(prediction, -1) seg_mask = tf.squeeze(seg_mask) return seg_mask def get_predictions(image: Image): predicted_segmentation_mask = run_model(image) preprocessed_image = preprocess_input(image) preprocessed_image = tf.squeeze(preprocessed_image, 0) pred_img = get_seg_overlay(preprocessed_image.numpy(), predicted_segmentation_mask.numpy()) return Image.fromarray(pred_img)import gradio as gr title = "Simple demo for a semantic segmentation model trained on the Sidewalks dataset." description = """ Note that the outputs obtained in this demo won't be state-of-the-art. The underlying project has a different objective focusing more on the ops side of deploying a semantic segmentation model. For more details, check out the repository: https://github.com/deep-diver/semantic-segmentation-ml-pipeline/. """ demo = gr.Interface( get_predictions, gr.inputs.Image(type="pil"), "pil", allow_flagging="never", title=title, description=description, examples=[["test-image1.png"], ["test-image2.png"]] ) demo.launch(debug=True)
deep-diver/semantic-segmentation-ml-pipeline
notebooks/inference_SavedModel_VertexEndpoint.ipynb
model_path = "model" !mkdir {model_path} !cp -r tmp/Format-Serving/* {model_path}/import tensorflow as tf model = tf.keras.models.load_model(model_path)test_image_path = tf.keras.utils.get_file( "test-image.png", "https://i.ibb.co/F58NjRq/test-image.png" ) test_label_path = tf.keras.utils.get_file( "gt-image.png", "https://i.ibb.co/BjPbS6c/gt-image.png" )import numpy as np from tensorflow.keras.applications import mobilenet_v2 with open(test_image_path, "rb") as f: test_image = f.read() with open(test_label_path, "rb") as f: gt_image = f.read() test_image = tf.io.decode_png(test_image, channels=3) test_image = tf.image.resize(test_image, size=(128, 128)) test_image = mobilenet_v2.preprocess_input(test_image) test_gt = tf.io.decode_png(gt_image, channels=1) input_numpy = np.array(test_image) input_tensor = tf.convert_to_tensor(input_numpy) input_tensor = input_tensor[tf.newaxis, ...]prediction = model.predict(input_tensor) print(prediction.shape) seg_mask = tf.math.argmax(prediction, -1) seg_mask = tf.squeeze(seg_mask) seg_maskimport matplotlib.pyplot as plt f, axs = plt.subplots(1, 2) f.set_figheight(10) f.set_figwidth(20) axs[0].set_title("Prediction", {"fontsize": 40}) axs[0].imshow(seg_mask) axs[0].axis("off") axs[1].set_title("Ground truth", {"fontsize": 40}) axs[1].imshow(tf.squeeze(test_gt, axis=2)) axs[1].axis("off") plt.show()import google.auth # pip install -U google-auth from google.auth.transport.requests import AuthorizedSession import tensorflow as tf import base64 import jsonPROJECT_ID = "gcp-ml-172005" REGION = "us-central1" ENDPOINT_ID = "2220630858061053952"credentials, _ = google.auth.default() service_endpoint = f"https://{REGION}-aiplatform.googleapis.com" authed_session = AuthorizedSession(credentials)url = "{}/v1/projects/{}/locations/{}/endpoints/{}:predict".format( service_endpoint, PROJECT_ID, REGION, ENDPOINT_ID ) print("Endpoint: ", url)serving_input = list( model.signatures["serving_default"].structured_input_signature[1].keys() )[0] print("Serving function input:", serving_input)with open(test_image_path, "rb") as f: image = f.read() b64str = base64.urlsafe_b64encode(image).decode('utf-8')single_instance_request_body = { "instances": [ {serving_input: b64str} ] } two_instances_request_body = { "instances": [ {serving_input: b64str}, {serving_input: b64str}, ] }response = authed_session.post(url, data=json.dumps(single_instance_request_body)) print(response) print(response.content)import ast response = response.content.decode('utf-8') response = ast.literal_eval(response) seg_mask_ve = tf.convert_to_tensor(response['predictions']) seg_mask_ve = tf.squeeze(seg_mask_ve) seg_mask_veimport matplotlib.pyplot as plt f, axs = plt.subplots(1, 3) f.set_figheight(10) f.set_figwidth(30) axs[0].set_title("Prediction(local)", {"fontsize": 40}) axs[0].imshow(seg_mask) axs[0].axis("off") axs[1].set_title("Prediction(vertex endpoint)", {"fontsize": 40}) axs[1].imshow(seg_mask_ve) axs[1].axis("off") axs[2].set_title("Ground truth", {"fontsize": 40}) axs[2].imshow(tf.squeeze(test_gt, axis=2)) axs[2].axis("off") plt.show()import numpy as np np.testing.assert_allclose(seg_mask, seg_mask_ve)response = authed_session.post(url, data=json.dumps(two_instances_request_body)) print(response) print(response.content)response = response.content.decode('utf-8') response = ast.literal_eval(response) seg_mask = tf.convert_to_tensor(response['predictions']) seg_mask = tf.squeeze(seg_mask) seg_mask
deep-diver/semantic-segmentation-ml-pipeline
notebooks/parse_tfrecords_pets.ipynb
import tensorflow as tfGCS_PATH = "gs://pets-tfrecords/pets-tfrecords" BATCH_SIZE = 4 AUTO = tf.data.AUTOTUNEdef parse_tfr(proto): feature_description = { "image": tf.io.VarLenFeature(tf.float32), "image_shape": tf.io.VarLenFeature(tf.int64), "label": tf.io.VarLenFeature(tf.float32), "label_shape": tf.io.VarLenFeature(tf.int64), } rec = tf.io.parse_single_example(proto, feature_description) image_shape = tf.sparse.to_dense(rec["image_shape"]) image = tf.reshape(tf.sparse.to_dense(rec["image"]), image_shape) label_shape = tf.sparse.to_dense(rec["label_shape"]) label = tf.reshape(tf.sparse.to_dense(rec["label"]), label_shape) return {"pixel_values": image, "label": label} def prepare_dataset(GCS_PATH=GCS_PATH, split="train", batch_size=BATCH_SIZE): if split not in ["train", "val"]: raise ValueError( "Invalid split provided. Supports splits are: `train` and `val`." ) dataset = tf.data.TFRecordDataset( [filename for filename in tf.io.gfile.glob(f"{GCS_PATH}/{split}-*")], num_parallel_reads=AUTO, ).map(parse_tfr, num_parallel_calls=AUTO) if split == "train": dataset = dataset.shuffle(batch_size * 2) dataset = dataset.batch(batch_size) dataset = dataset.prefetch(AUTO) return datasettrain_dataset = prepare_dataset() val_dataset = prepare_dataset(split="val")for batch in train_dataset.take(1): print(batch["pixel_values"].shape, batch["label"].shape)for batch in val_dataset.take(1): print(batch["pixel_values"].shape, batch["label"].shape)
deep-diver/semantic-segmentation-ml-pipeline
notebooks/parse_tfrecords_sidewalks.ipynb
import tensorflow as tfGCS_PATH_FULL_RESOUTION = "gs://sidewalks-tfx-fullres/sidewalks-tfrecords" GCS_PATH_LOW_RESOLUTION = "gs://sidewalks-tfx-lowres/sidewalks-tfrecords" BATCH_SIZE = 4 AUTO = tf.data.AUTOTUNEdef parse_tfr(proto): feature_description = { "image": tf.io.VarLenFeature(tf.float32), "image_shape": tf.io.VarLenFeature(tf.int64), "label": tf.io.VarLenFeature(tf.float32), "label_shape": tf.io.VarLenFeature(tf.int64), } rec = tf.io.parse_single_example(proto, feature_description) image_shape = tf.sparse.to_dense(rec["image_shape"]) image = tf.reshape(tf.sparse.to_dense(rec["image"]), image_shape) label_shape = tf.sparse.to_dense(rec["label_shape"]) label = tf.reshape(tf.sparse.to_dense(rec["label"]), label_shape) return {"pixel_values": image, "label": label} def prepare_dataset(GCS_PATH=GCS_PATH_FULL_RESOUTION, split="train", batch_size=BATCH_SIZE): if split not in ["train", "val"]: raise ValueError( "Invalid split provided. Supports splits are: `train` and `val`." ) dataset = tf.data.TFRecordDataset( [filename for filename in tf.io.gfile.glob(f"{GCS_PATH}/{split}-*")], num_parallel_reads=AUTO, ).map(parse_tfr, num_parallel_calls=AUTO) if split == "train": dataset = dataset.shuffle(batch_size * 2) dataset = dataset.batch(batch_size) dataset = dataset.prefetch(AUTO) return datasettrain_dataset = prepare_dataset() val_dataset = prepare_dataset(split="val")for batch in train_dataset.take(1): print(batch["pixel_values"].shape, batch["label"].shape)for batch in val_dataset.take(1): print(batch["pixel_values"].shape, batch["label"].shape)train_dataset = prepare_dataset(GCS_PATH_LOW_RESOLUTION) val_dataset = prepare_dataset(GCS_PATH_LOW_RESOLUTION, split="val")for batch in train_dataset.take(1): print(batch["pixel_values"].shape, batch["label"].shape)for batch in val_dataset.take(1): print(batch["pixel_values"].shape, batch["label"].shape)
deep-diver/semantic-segmentation-ml-pipeline
notebooks/tfx_pipeline_pets.ipynb
data_path = "gs://sidewalks-tfx-lowres/sidewalks-tfrecords" local_data_path = "data" model_file = "modules/model.py" model_fn = "modules.model.run_fn" preprocessing_file = "modules/preprocessing.py" preprocessing_fn = "modules.preprocessing.preprocessing_fn"import tfx tfx.__version__from tfx import v1 as tfx from tfx.orchestration.experimental.interactive.interactive_context import InteractiveContext from tfx.components import ImportExampleGen from tfx.components import SchemaGen from tfx.components import StatisticsGen from tfx.components import Trainer from tfx.components import Transform from tfx.components import Evaluator from tfx.proto import example_gen_pb2 from tfx.types import Channel from tfx.types.standard_artifacts import Model from tfx.types.standard_artifacts import ModelBlessing from tfx.dsl.components.common import resolver from tfx.dsl.experimental.latest_blessed_model_resolver import LatestBlessedModelResolver import tensorflow_model_analysis as tfmacontext = InteractiveContext()input_config = example_gen_pb2.Input( splits=[ example_gen_pb2.Input.Split(name="train", pattern="train-*.tfrec"), example_gen_pb2.Input.Split(name="eval", pattern="val-*.tfrec"), ] ) example_gen = ImportExampleGen( input_base=local_data_path, input_config=input_config )context.run(example_gen)statistics_gen = StatisticsGen(examples=example_gen.outputs["examples"])context.run(statistics_gen)context.show(statistics_gen.outputs['statistics'])schema_gen = SchemaGen(statistics=statistics_gen.outputs['statistics'])context.run(schema_gen)%%writefile {preprocessing_file} import tensorflow as tf from tensorflow.keras.applications import mobilenet_v2 _INPUT_IMG_SIZE = 128 _IMAGE_KEY = "image" _IMAGE_SHAPE_KEY = "image_shape" _LABEL_KEY = "label" _LABEL_SHAPE_KEY = "label_shape" def _transformed_name(key: str) -> str: return key + "_xf" # output should have the same keys as inputs def preprocess(inputs): image_shape = inputs[_IMAGE_SHAPE_KEY] label_shape = inputs[_LABEL_SHAPE_KEY] images = tf.reshape(inputs[_IMAGE_KEY], [image_shape[0], image_shape[1], 3]) labels = tf.reshape(inputs[_LABEL_KEY], [label_shape[0], label_shape[1], 1]) return { _IMAGE_KEY: images, _IMAGE_SHAPE_KEY: inputs[_IMAGE_SHAPE_KEY], _LABEL_KEY: labels, _LABEL_SHAPE_KEY: inputs[_LABEL_SHAPE_KEY], } def preprocessing_fn(inputs): """tf.transform's callback function for preprocessing inputs. Args: inputs: map from feature keys to raw not-yet-transformed features. Returns: Map from string feature key to transformed feature operations. """ # print(inputs) outputs = {} features = tf.map_fn(preprocess, inputs) features[_IMAGE_KEY] = tf.image.resize(features[_IMAGE_KEY], [_INPUT_IMG_SIZE, _INPUT_IMG_SIZE]) features[_LABEL_KEY] = tf.image.resize(features[_LABEL_KEY], [_INPUT_IMG_SIZE, _INPUT_IMG_SIZE]) image_features = mobilenet_v2.preprocess_input(features[_IMAGE_KEY]) outputs[_transformed_name(_IMAGE_KEY)] = image_features outputs[_transformed_name(_LABEL_KEY)] = features[_LABEL_KEY] return outputstransform = Transform( examples=example_gen.outputs["examples"], schema=schema_gen.outputs["schema"], preprocessing_fn=preprocessing_fn, )context.run(transform)%%writefile {model_file} from typing import List, Dict, Tuple import absl import tensorflow as tf import tensorflow_transform as tft from tensorflow.keras.optimizers import Adam from tfx_bsl.tfxio import dataset_options from tfx.components.trainer.fn_args_utils import FnArgs from tfx.components.trainer.fn_args_utils import DataAccessor _CONCRETE_INPUT = "pixel_values" _RAW_IMG_SIZE = 256 _INPUT_IMG_SIZE = 128 _TRAIN_LENGTH = 800 _EVAL_LENGTH = 200 _TRAIN_BATCH_SIZE = 64 _EVAL_BATCH_SIZE = 64 _EPOCHS = 1 _LR = 0.00006 _IMAGE_KEY = "image" _LABEL_KEY = "label" def INFO(text: str): absl.logging.info(text) def _transformed_name(key: str) -> str: return key + "_xf" """ _serving_preprocess, _serving_preprocess_fn, and _model_exporter functions are defined to provide pre- processing capabilities when the model is served. """ def _serving_preprocess(string_input): decoded_input = tf.io.decode_base64(string_input) decoded = tf.io.decode_jpeg(decoded_input, channels=3) decoded = decoded / 255 resized = tf.image.resize(decoded, size=(_INPUT_IMG_SIZE, _INPUT_IMG_SIZE)) return resized @tf.function(input_signature=[tf.TensorSpec([None], tf.string)]) def _serving_preprocess_fn(string_input): decoded_images = tf.map_fn( _serving_preprocess, string_input, dtype=tf.float32, back_prop=False ) return {_CONCRETE_INPUT: decoded_images} def _model_exporter(model: tf.keras.Model): m_call = tf.function(model.call).get_concrete_function( tf.TensorSpec(shape=[None, _INPUT_IMG_SIZE, _INPUT_IMG_SIZE, 3], dtype=tf.float32, name=_CONCRETE_INPUT) ) @tf.function(input_signature=[tf.TensorSpec([None], tf.string)]) def serving_fn(string_input): images = _serving_preprocess_fn(string_input) logits = m_call(**images) seg_mask = tf.math.argmax(logits, -1) return {"seg_mask": seg_mask} return serving_fn def _get_transform_features_signature(model, tf_transform_output): # the layer is added as an attribute to the model in order to make sure that # the model assets are handled correctly when exporting. model.tft_layer = tf_transform_output.transform_features_layer() @tf.function( input_signature=[tf.TensorSpec(shape=[None], dtype=tf.string, name="examples")] ) def serve_tf_examples_fn(serialized_tf_examples): """Returns the output to be used in the serving signature.""" feature_spec = tf_transform_output.raw_feature_spec() parsed_features = tf.io.parse_example(serialized_tf_examples, feature_spec) transformed_features = model.tft_layer(parsed_features) return transformed_features return serve_tf_examples_fn def _get_tf_examples_serving_signature(model, tf_transform_output): """ Returns a serving signature that accepts `tensorflow.Example`. This signature will be used for evaluation or bulk inference. """ model.tft_layer = tf_transform_output.transform_features_layer() @tf.function( input_signature=[tf.TensorSpec(shape=[None], dtype=tf.string, name="examples")] ) def serve_tf_examples_fn( serialized_tf_example: tf.Tensor, ) -> Dict[str, tf.Tensor]: """Returns the output to be used in the serving signature.""" # Load the schema of raw examples. raw_feature_spec = tf_transform_output.raw_feature_spec() # Remove label feature since these will not be present at serving time. # Parse the examples using schema into raw features raw_features = tf.io.parse_example(serialized_tf_example, raw_feature_spec) # Preprocess the raw features transformed_features = model.tft_layer(raw_features) # Run preprocessed inputs through the model to get the prediction outputs = model(transformed_features) return { _transformed_name(_LABEL_KEY): outputs } return serve_tf_examples_fn """ _input_fn reads TFRecord files passed from the upstream TFX component, Transform. Assume the dataset is already transformed appropriately. """ def _input_fn( file_pattern: List[str], data_accessor: DataAccessor, tf_transform_output: tft.TFTransformOutput, is_train: bool = False, batch_size: int = 200, ) -> tf.data.Dataset: dataset = data_accessor.tf_dataset_factory( file_pattern, dataset_options.TensorFlowDatasetOptions( batch_size=batch_size, label_key=_transformed_name(_LABEL_KEY) ), tf_transform_output.transformed_metadata.schema, ) return dataset def _build_model(num_labels) -> tf.keras.Model: base_model = tf.keras.applications.MobileNetV2( input_shape=[128, 128, 3], include_top=False ) # Use the activations of these layers layer_names = [ "block_1_expand_relu", # 64x64 "block_3_expand_relu", # 32x32 "block_6_expand_relu", # 16x16 "block_13_expand_relu", # 8x8 "block_16_project", # 4x4 ] base_model_outputs = [base_model.get_layer(name).output for name in layer_names] # Create the feature extraction model down_stack = tf.keras.Model(inputs=base_model.input, outputs=base_model_outputs) down_stack.trainable = False up_stack = [ upsample(512, 3), # 4x4 -> 8x8 upsample(256, 3), # 8x8 -> 16x16 upsample(128, 3), # 16x16 -> 32x32 upsample(64, 3), # 32x32 -> 64x64 ] inputs = tf.keras.layers.Input( shape=[128, 128, 3], name=_transformed_name(_IMAGE_KEY) ) # Downsampling through the model skips = down_stack(inputs) x = skips[-1] skips = reversed(skips[:-1]) # Upsampling and establishing the skip connections for up, skip in zip(up_stack, skips): x = up(x) concat = tf.keras.layers.Concatenate() x = concat([x, skip]) # This is the last layer of the model last = tf.keras.layers.Conv2DTranspose( filters=num_labels, kernel_size=3, strides=2, padding="same", name=_transformed_name(_LABEL_KEY) ) # 64x64 -> 128x128 x = last(x) model = tf.keras.Model(inputs=inputs, outputs=x) model.compile( optimizer=Adam(learning_rate=_LR), loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True), metrics=["sparse_categorical_accuracy"], ) return model """ InstanceNormalization class and upsample function are borrowed from pix2pix in [TensorFlow Example repository]( https://github.com/tensorflow/examples/tree/master/tensorflow_examples/models/pix2pix) """ class InstanceNormalization(tf.keras.layers.Layer): """Instance Normalization Layer (https://arxiv.org/abs/1607.08022).""" def __init__(self, epsilon=1e-5): super(InstanceNormalization, self).__init__() self.epsilon = epsilon def build(self, input_shape): self.scale = self.add_weight( name="scale", shape=input_shape[-1:], initializer=tf.random_normal_initializer(1.0, 0.02), trainable=True, ) self.offset = self.add_weight( name="offset", shape=input_shape[-1:], initializer="zeros", trainable=True ) def call(self, x): mean, variance = tf.nn.moments(x, axes=[1, 2], keepdims=True) inv = tf.math.rsqrt(variance + self.epsilon) normalized = (x - mean) * inv return self.scale * normalized + self.offset def upsample(filters, size, norm_type="batchnorm", apply_dropout=False): """Upsamples an input. Conv2DTranspose => Batchnorm => Dropout => Relu Args: filters: number of filters size: filter size norm_type: Normalization type; either 'batchnorm' or 'instancenorm'. apply_dropout: If True, adds the dropout layer Returns: Upsample Sequential Model """ initializer = tf.random_normal_initializer(0.0, 0.02) result = tf.keras.Sequential() result.add( tf.keras.layers.Conv2DTranspose( filters, size, strides=2, padding="same", kernel_initializer=initializer, use_bias=False, ) ) if norm_type.lower() == "batchnorm": result.add(tf.keras.layers.BatchNormalization()) elif norm_type.lower() == "instancenorm": result.add(InstanceNormalization()) if apply_dropout: result.add(tf.keras.layers.Dropout(0.5)) result.add(tf.keras.layers.ReLU()) return result def run_fn(fn_args: FnArgs): tf_transform_output = tft.TFTransformOutput(fn_args.transform_output) train_dataset = _input_fn( fn_args.train_files, fn_args.data_accessor, tf_transform_output, is_train=True, batch_size=_TRAIN_BATCH_SIZE, ) eval_dataset = _input_fn( fn_args.eval_files, fn_args.data_accessor, tf_transform_output, is_train=False, batch_size=_EVAL_BATCH_SIZE, ) num_labels = 35 model = _build_model(num_labels) model.fit( train_dataset, steps_per_epoch=_TRAIN_LENGTH // _TRAIN_BATCH_SIZE, validation_data=eval_dataset, validation_steps=_EVAL_LENGTH // _TRAIN_BATCH_SIZE, epochs=_EPOCHS, ) model.save( fn_args.serving_model_dir, save_format="tf", signatures={ "serving_default": _model_exporter(model), "transform_features": _get_transform_features_signature( model, tf_transform_output ), "from_examples": _get_tf_examples_serving_signature( model, tf_transform_output ), } ) trainer = Trainer( run_fn=model_fn, transformed_examples=transform.outputs["transformed_examples"], transform_graph=transform.outputs["transform_graph"], schema=schema_gen.outputs["schema"], )context.run(trainer)
deep-diver/semantic-segmentation-ml-pipeline
notebooks/tfx_pipeline_sidewalks.ipynb
data_path = "gs://sidewalks-tfx-lowres/sidewalks-tfrecords" local_data_path = "data" model_file = "modules/model.py" model_fn = "modules.model.run_fn" preprocessing_file = "modules/preprocessing.py" preprocessing_fn = "modules.preprocessing.preprocessing_fn"import tfx tfx.__version__from tfx import v1 as tfx from tfx.orchestration.experimental.interactive.interactive_context import InteractiveContext from tfx.components import ImportExampleGen from tfx.components import SchemaGen from tfx.components import StatisticsGen from tfx.components import Trainer from tfx.components import Transform from tfx.components import Evaluator from tfx.proto import example_gen_pb2 from tfx.types import Channel from tfx.types.standard_artifacts import Model from tfx.types.standard_artifacts import ModelBlessing from tfx.dsl.components.common import resolver from tfx.dsl.experimental.latest_blessed_model_resolver import LatestBlessedModelResolver import tensorflow_model_analysis as tfmacontext = InteractiveContext()input_config = example_gen_pb2.Input( splits=[ example_gen_pb2.Input.Split(name="train", pattern="train-*.tfrec"), example_gen_pb2.Input.Split(name="eval", pattern="val-*.tfrec"), ] ) example_gen = ImportExampleGen( input_base=local_data_path, input_config=input_config )context.run(example_gen)statistics_gen = StatisticsGen(examples=example_gen.outputs["examples"])context.run(statistics_gen)context.show(statistics_gen.outputs['statistics'])schema_gen = SchemaGen(statistics=statistics_gen.outputs['statistics'])context.run(schema_gen)%%writefile {preprocessing_file} import tensorflow as tf from tensorflow.keras.applications import mobilenet_v2 _INPUT_IMG_SIZE = 128 _IMAGE_KEY = "image" _IMAGE_SHAPE_KEY = "image_shape" _LABEL_KEY = "label" _LABEL_SHAPE_KEY = "label_shape" def _transformed_name(key: str) -> str: return key + "_xf" # output should have the same keys as inputs def preprocess(inputs): image_shape = inputs[_IMAGE_SHAPE_KEY] label_shape = inputs[_LABEL_SHAPE_KEY] images = tf.reshape(inputs[_IMAGE_KEY], [image_shape[0], image_shape[1], 3]) labels = tf.reshape(inputs[_LABEL_KEY], [label_shape[0], label_shape[1], 1]) return { _IMAGE_KEY: images, _IMAGE_SHAPE_KEY: inputs[_IMAGE_SHAPE_KEY], _LABEL_KEY: labels, _LABEL_SHAPE_KEY: inputs[_LABEL_SHAPE_KEY], } def preprocessing_fn(inputs): """tf.transform's callback function for preprocessing inputs. Args: inputs: map from feature keys to raw not-yet-transformed features. Returns: Map from string feature key to transformed feature operations. """ # print(inputs) outputs = {} features = tf.map_fn(preprocess, inputs) features[_IMAGE_KEY] = tf.image.resize(features[_IMAGE_KEY], [_INPUT_IMG_SIZE, _INPUT_IMG_SIZE]) features[_LABEL_KEY] = tf.image.resize(features[_LABEL_KEY], [_INPUT_IMG_SIZE, _INPUT_IMG_SIZE]) image_features = mobilenet_v2.preprocess_input(features[_IMAGE_KEY]) outputs[_transformed_name(_IMAGE_KEY)] = image_features outputs[_transformed_name(_LABEL_KEY)] = features[_LABEL_KEY] return outputstransform = Transform( examples=example_gen.outputs["examples"], schema=schema_gen.outputs["schema"], preprocessing_fn=preprocessing_fn, )context.run(transform)%%writefile {model_file} from typing import List, Dict, Tuple import absl import tensorflow as tf import tensorflow_transform as tft from tensorflow.keras.optimizers import Adam from tfx_bsl.tfxio import dataset_options from tfx.components.trainer.fn_args_utils import FnArgs from tfx.components.trainer.fn_args_utils import DataAccessor _CONCRETE_INPUT = "pixel_values" _RAW_IMG_SIZE = 256 _INPUT_IMG_SIZE = 128 _TRAIN_LENGTH = 800 _EVAL_LENGTH = 200 _TRAIN_BATCH_SIZE = 64 _EVAL_BATCH_SIZE = 64 _EPOCHS = 1 _LR = 0.00006 _IMAGE_KEY = "image" _LABEL_KEY = "label" def INFO(text: str): absl.logging.info(text) def _transformed_name(key: str) -> str: return key + "_xf" """ _serving_preprocess, _serving_preprocess_fn, and _model_exporter functions are defined to provide pre- processing capabilities when the model is served. """ def _serving_preprocess(string_input): decoded_input = tf.io.decode_base64(string_input) decoded = tf.io.decode_jpeg(decoded_input, channels=3) decoded = decoded / 255 resized = tf.image.resize(decoded, size=(_INPUT_IMG_SIZE, _INPUT_IMG_SIZE)) return resized @tf.function(input_signature=[tf.TensorSpec([None], tf.string)]) def _serving_preprocess_fn(string_input): decoded_images = tf.map_fn( _serving_preprocess, string_input, dtype=tf.float32, back_prop=False ) return {_CONCRETE_INPUT: decoded_images} def _model_exporter(model: tf.keras.Model): m_call = tf.function(model.call).get_concrete_function( tf.TensorSpec(shape=[None, _INPUT_IMG_SIZE, _INPUT_IMG_SIZE, 3], dtype=tf.float32, name=_CONCRETE_INPUT) ) @tf.function(input_signature=[tf.TensorSpec([None], tf.string)]) def serving_fn(string_input): images = _serving_preprocess_fn(string_input) logits = m_call(**images) seg_mask = tf.math.argmax(logits, -1) return {"seg_mask": seg_mask} return serving_fn def _get_transform_features_signature(model, tf_transform_output): # the layer is added as an attribute to the model in order to make sure that # the model assets are handled correctly when exporting. model.tft_layer = tf_transform_output.transform_features_layer() @tf.function( input_signature=[tf.TensorSpec(shape=[None], dtype=tf.string, name="examples")] ) def serve_tf_examples_fn(serialized_tf_examples): """Returns the output to be used in the serving signature.""" feature_spec = tf_transform_output.raw_feature_spec() parsed_features = tf.io.parse_example(serialized_tf_examples, feature_spec) transformed_features = model.tft_layer(parsed_features) return transformed_features return serve_tf_examples_fn def _get_tf_examples_serving_signature(model, tf_transform_output): """ Returns a serving signature that accepts `tensorflow.Example`. This signature will be used for evaluation or bulk inference. """ model.tft_layer = tf_transform_output.transform_features_layer() @tf.function( input_signature=[tf.TensorSpec(shape=[None], dtype=tf.string, name="examples")] ) def serve_tf_examples_fn( serialized_tf_example: tf.Tensor, ) -> Dict[str, tf.Tensor]: """Returns the output to be used in the serving signature.""" # Load the schema of raw examples. raw_feature_spec = tf_transform_output.raw_feature_spec() # Remove label feature since these will not be present at serving time. # Parse the examples using schema into raw features raw_features = tf.io.parse_example(serialized_tf_example, raw_feature_spec) # Preprocess the raw features transformed_features = model.tft_layer(raw_features) # Run preprocessed inputs through the model to get the prediction outputs = model(transformed_features) return { _transformed_name(_LABEL_KEY): outputs } return serve_tf_examples_fn """ _input_fn reads TFRecord files passed from the upstream TFX component, Transform. Assume the dataset is already transformed appropriately. """ def _input_fn( file_pattern: List[str], data_accessor: DataAccessor, tf_transform_output: tft.TFTransformOutput, is_train: bool = False, batch_size: int = 200, ) -> tf.data.Dataset: dataset = data_accessor.tf_dataset_factory( file_pattern, dataset_options.TensorFlowDatasetOptions( batch_size=batch_size, label_key=_transformed_name(_LABEL_KEY) ), tf_transform_output.transformed_metadata.schema, ) return dataset def _build_model(num_labels) -> tf.keras.Model: base_model = tf.keras.applications.MobileNetV2( input_shape=[128, 128, 3], include_top=False ) # Use the activations of these layers layer_names = [ "block_1_expand_relu", # 64x64 "block_3_expand_relu", # 32x32 "block_6_expand_relu", # 16x16 "block_13_expand_relu", # 8x8 "block_16_project", # 4x4 ] base_model_outputs = [base_model.get_layer(name).output for name in layer_names] # Create the feature extraction model down_stack = tf.keras.Model(inputs=base_model.input, outputs=base_model_outputs) down_stack.trainable = False up_stack = [ upsample(512, 3), # 4x4 -> 8x8 upsample(256, 3), # 8x8 -> 16x16 upsample(128, 3), # 16x16 -> 32x32 upsample(64, 3), # 32x32 -> 64x64 ] inputs = tf.keras.layers.Input( shape=[128, 128, 3], name=_transformed_name(_IMAGE_KEY) ) # Downsampling through the model skips = down_stack(inputs) x = skips[-1] skips = reversed(skips[:-1]) # Upsampling and establishing the skip connections for up, skip in zip(up_stack, skips): x = up(x) concat = tf.keras.layers.Concatenate() x = concat([x, skip]) # This is the last layer of the model last = tf.keras.layers.Conv2DTranspose( filters=num_labels, kernel_size=3, strides=2, padding="same", name=_transformed_name(_LABEL_KEY) ) # 64x64 -> 128x128 x = last(x) model = tf.keras.Model(inputs=inputs, outputs=x) model.compile( optimizer=Adam(learning_rate=_LR), loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True), metrics=["sparse_categorical_accuracy"], ) return model """ InstanceNormalization class and upsample function are borrowed from pix2pix in [TensorFlow Example repository]( https://github.com/tensorflow/examples/tree/master/tensorflow_examples/models/pix2pix) """ class InstanceNormalization(tf.keras.layers.Layer): """Instance Normalization Layer (https://arxiv.org/abs/1607.08022).""" def __init__(self, epsilon=1e-5): super(InstanceNormalization, self).__init__() self.epsilon = epsilon def build(self, input_shape): self.scale = self.add_weight( name="scale", shape=input_shape[-1:], initializer=tf.random_normal_initializer(1.0, 0.02), trainable=True, ) self.offset = self.add_weight( name="offset", shape=input_shape[-1:], initializer="zeros", trainable=True ) def call(self, x): mean, variance = tf.nn.moments(x, axes=[1, 2], keepdims=True) inv = tf.math.rsqrt(variance + self.epsilon) normalized = (x - mean) * inv return self.scale * normalized + self.offset def upsample(filters, size, norm_type="batchnorm", apply_dropout=False): """Upsamples an input. Conv2DTranspose => Batchnorm => Dropout => Relu Args: filters: number of filters size: filter size norm_type: Normalization type; either 'batchnorm' or 'instancenorm'. apply_dropout: If True, adds the dropout layer Returns: Upsample Sequential Model """ initializer = tf.random_normal_initializer(0.0, 0.02) result = tf.keras.Sequential() result.add( tf.keras.layers.Conv2DTranspose( filters, size, strides=2, padding="same", kernel_initializer=initializer, use_bias=False, ) ) if norm_type.lower() == "batchnorm": result.add(tf.keras.layers.BatchNormalization()) elif norm_type.lower() == "instancenorm": result.add(InstanceNormalization()) if apply_dropout: result.add(tf.keras.layers.Dropout(0.5)) result.add(tf.keras.layers.ReLU()) return result def run_fn(fn_args: FnArgs): tf_transform_output = tft.TFTransformOutput(fn_args.transform_output) train_dataset = _input_fn( fn_args.train_files, fn_args.data_accessor, tf_transform_output, is_train=True, batch_size=_TRAIN_BATCH_SIZE, ) eval_dataset = _input_fn( fn_args.eval_files, fn_args.data_accessor, tf_transform_output, is_train=False, batch_size=_EVAL_BATCH_SIZE, ) num_labels = 35 model = _build_model(num_labels) model.fit( train_dataset, steps_per_epoch=_TRAIN_LENGTH // _TRAIN_BATCH_SIZE, validation_data=eval_dataset, validation_steps=_EVAL_LENGTH // _TRAIN_BATCH_SIZE, epochs=_EPOCHS, ) model.save( fn_args.serving_model_dir, save_format="tf", signatures={ "serving_default": _model_exporter(model), "transform_features": _get_transform_features_signature( model, tf_transform_output ), "from_examples": _get_tf_examples_serving_signature( model, tf_transform_output ), } ) trainer = Trainer( run_fn=model_fn, transformed_examples=transform.outputs["transformed_examples"], transform_graph=transform.outputs["transform_graph"], schema=schema_gen.outputs["schema"], )context.run(trainer)
deep-diver/semantic-segmentation-ml-pipeline
notebooks/unet_training_sidewalks.ipynb
import tensorflow as tf import numpy as npGCS_PATH = "gs://sidewalks-tfx-hf/sidewalks-tfrecords" BATCH_SIZE = 2 AUTO = tf.data.AUTOTUNEdef parse_tfr(proto): feature_description = { "image": tf.io.FixedLenFeature([], tf.string), "label": tf.io.FixedLenFeature([], tf.string) } rec = tf.io.parse_single_example(proto, feature_description) image = tf.io.parse_tensor(rec["image"], tf.float32) label = tf.io.parse_tensor(rec["label"], tf.float32) return {"pixel_values": image, "labels": label} def prepare_dataset(split="train", batch_size=BATCH_SIZE): if split not in ["train", "val"]: raise ValueError( "Invalid split provided. Supports splits are: `train` and `val`." ) dataset = tf.data.TFRecordDataset( [filename for filename in tf.io.gfile.glob(f"{GCS_PATH}/{split}-*")], num_parallel_reads=AUTO, ).map(parse_tfr, num_parallel_calls=AUTO) if split == "train": dataset = dataset.shuffle(batch_size * 2) dataset = dataset.batch(batch_size) dataset = dataset.prefetch(AUTO) return datasettrain_dataset = prepare_dataset() val_dataset = prepare_dataset(split="val")batch = next(iter(train_dataset)) batch[0].shape, batch[1].shapedef preprocess(example_batch): images = example_batch["pixel_values"] images = tf.transpose(images, perm=[0, 1, 2, 3]) # (batch_size, height, width, num_channels) labels = tf.expand_dims(example_batch["labels"], -1) # Adds extra dimension, otherwise tf.image.resize won't work. labels = tf.transpose(labels, perm=[0, 1, 2, 3]) # So, that TF can evaluation the shapes. images = tf.image.resize(images, (128, 128)) labels = tf.image.resize(labels, (128, 128)) # images = tf.transpose(images, perm=[0, 3, 1, 2]) # (batch_size, num_channels, height, width) labels = tf.squeeze(labels, -1) return images, labelstrain_dataset = train_dataset.map(preprocess) val_dataset = val_dataset.map(preprocess)# Investigate a single batch. batch = next(iter(train_dataset)) batch[0].shape, batch[1].shapedef sidewalk_palette(): """Sidewalk palette that maps each class to RGB values.""" return [ [0, 0, 0], [216, 82, 24], [255, 255, 0], [125, 46, 141], [118, 171, 47], [161, 19, 46], [255, 0, 0], [0, 128, 128], [190, 190, 0], [0, 255, 0], [0, 0, 255], [170, 0, 255], [84, 84, 0], [84, 170, 0], [84, 255, 0], [170, 84, 0], [170, 170, 0], [170, 255, 0], [255, 84, 0], [255, 170, 0], [255, 255, 0], [33, 138, 200], [0, 170, 127], [0, 255, 127], [84, 0, 127], [84, 84, 127], [84, 170, 127], [84, 255, 127], [170, 0, 127], [170, 84, 127], [170, 170, 127], [170, 255, 127], [255, 0, 127], [255, 84, 127], [255, 170, 127], ]def get_seg_overlay(image, seg): # image = tf.transpose(image, [1, 2, 0]) color_seg = np.zeros( (seg.shape[0], seg.shape[1], 3), dtype=np.uint8 ) # height, width, 3 palette = np.array(sidewalk_palette()) for label, color in enumerate(palette): color_seg[seg == label, :] = color # Show image + mask img = np.array(image) * 0.5 + color_seg * 0.5 img = img.astype(np.uint8) return imggt_img = get_seg_overlay( batch[0][0], np.array(batch[1][0]) )import matplotlib.pyplot as plt f, axs = plt.subplots(1, 1) f.set_figheight(5) f.set_figwidth(10) axs.set_title("Ground truth", {"fontsize": 20}) axs.imshow(gt_img) axs.axis("off") plt.show()class InstanceNormalization(tf.keras.layers.Layer): """Instance Normalization Layer (https://arxiv.org/abs/1607.08022).""" def __init__(self, epsilon=1e-5): super(InstanceNormalization, self).__init__() self.epsilon = epsilon def build(self, input_shape): self.scale = self.add_weight( name='scale', shape=input_shape[-1:], initializer=tf.random_normal_initializer(1., 0.02), trainable=True) self.offset = self.add_weight( name='offset', shape=input_shape[-1:], initializer='zeros', trainable=True) def call(self, x): mean, variance = tf.nn.moments(x, axes=[1, 2], keepdims=True) inv = tf.math.rsqrt(variance + self.epsilon) normalized = (x - mean) * inv return self.scale * normalized + self.offsetdef upsample(filters, size, norm_type='batchnorm', apply_dropout=False): """Upsamples an input. Conv2DTranspose => Batchnorm => Dropout => Relu Args: filters: number of filters size: filter size norm_type: Normalization type; either 'batchnorm' or 'instancenorm'. apply_dropout: If True, adds the dropout layer Returns: Upsample Sequential Model """ initializer = tf.random_normal_initializer(0., 0.02) result = tf.keras.Sequential() result.add( tf.keras.layers.Conv2DTranspose(filters, size, strides=2, padding='same', kernel_initializer=initializer, use_bias=False)) if norm_type.lower() == 'batchnorm': result.add(tf.keras.layers.BatchNormalization()) elif norm_type.lower() == 'instancenorm': result.add(InstanceNormalization()) if apply_dropout: result.add(tf.keras.layers.Dropout(0.5)) result.add(tf.keras.layers.ReLU()) return resultbase_model = tf.keras.applications.MobileNetV2(input_shape=[128, 128, 3], include_top=False) # Use the activations of these layers layer_names = [ 'block_1_expand_relu', # 64x64 'block_3_expand_relu', # 32x32 'block_6_expand_relu', # 16x16 'block_13_expand_relu', # 8x8 'block_16_project', # 4x4 ] base_model_outputs = [base_model.get_layer(name).output for name in layer_names] # Create the feature extraction model down_stack = tf.keras.Model(inputs=base_model.input, outputs=base_model_outputs) down_stack.trainable = Falseup_stack = [ upsample(512, 3), # 4x4 -> 8x8 upsample(256, 3), # 8x8 -> 16x16 upsample(128, 3), # 16x16 -> 32x32 upsample(64, 3), # 32x32 -> 64x64 ]def unet_model(output_channels:int): inputs = tf.keras.layers.Input(shape=[128, 128, 3], name="pixel_values") # Downsampling through the model skips = down_stack(inputs) x = skips[-1] skips = reversed(skips[:-1]) # Upsampling and establishing the skip connections for up, skip in zip(up_stack, skips): x = up(x) concat = tf.keras.layers.Concatenate() x = concat([x, skip]) # This is the last layer of the model last = tf.keras.layers.Conv2DTranspose( filters=output_channels, kernel_size=3, strides=2, padding='same', name="labels") #64x64 -> 128x128 x = last(x) return tf.keras.Model(inputs=inputs, outputs=x)OUTPUT_CLASSES = len(sidewalk_palette()) model = unet_model(output_channels=OUTPUT_CLASSES) model.compile(optimizer='adam', loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True), metrics=['accuracy'])model.summary()EPOCHS = 5 model_history = model.fit( train_dataset, validation_data=val_dataset, epochs=EPOCHS )test_batch = next(iter(val_dataset))pred_mask = model.predict(test_batch[0])pred_seg = tf.math.argmax(pred_mask[0], axis=-1)import matplotlib.pyplot as plt seg_im = get_seg_overlay(test_batch[0][0], pred_seg) f, axis = plt.subplots(1, 2) f.set_figheight(5) f.set_figwidth(20) axis[0].set_title("Image", {"fontsize": 20}) axis[1].set_title("Prediction", {"fontsize": 20}) axis[0].axis("off") axis[1].axis("off") axis[0].imshow(test_batch[0][0]) axis[1].imshow(seg_im)_IMAGE_SHAPE = (128, 128) _CONCRETE_INPUT = "pixel_values" def _serving_preprocess(string_input): decoded_input = tf.io.decode_base64(string_input) decoded = tf.io.decode_jpeg(decoded_input, channels=3) decoded = decoded / 255 resized = tf.image.resize(decoded, size=_IMAGE_SHAPE) return resized @tf.function(input_signature=[tf.TensorSpec([None], tf.string)]) def _serving_preprocess_fn(string_input): decoded_images = tf.map_fn( _serving_preprocess, string_input, dtype=tf.float32, back_prop=False ) return {_CONCRETE_INPUT: decoded_images} def _model_exporter(model: tf.keras.Model): m_call = tf.function(model.call).get_concrete_function( tf.TensorSpec( shape=[None, 128, 128, 3], dtype=tf.float32, name=_CONCRETE_INPUT ) ) @tf.function(input_signature=[tf.TensorSpec([None], tf.string)]) def serving_fn(string_input): images = _serving_preprocess_fn(string_input) logits = m_call(**images) seg_mask = tf.math.argmax(logits, -1) return {"seg_mask": seg_mask} return serving_fnmodel.save( "./test_model", save_format="tf", signatures=_model_exporter(model) )model = tf.keras.models.load_model('./test_model')pred_mask = model.predict(test_batch[0])pred_mask.shape
deep-diver/semantic-segmentation-ml-pipeline
tfrecords/create_tfrecords.py
""" Script to generate TFRecord shards from the Sidewalks dataset as shown in this blog post: https://huggingface.co/blog/fine-tune-segformer. The recommended way to obtain TFRecord shards is via an Apache Beam Pipeline with an execution runner like Dataflow. Example: https://github.com/GoogleCloudPlatform/practical-ml-vision-book/blob/master/05_create_dataset/jpeg_to_tfrecord.py. Usage: python create_tfrecords --batch_size 16 python create_tfrecords --resize 256 # without --resize flag, no resizing is applied References: * https://github.com/GoogleCloudPlatform/practical-ml-vision-book/blob/master/05_create_dataset/05_split_tfrecord.ipynb * https://www.tensorflow.org/tutorials/images/segmentation """ import argparse import math import os from typing import Tuple import datasets import numpy as np import tensorflow as tf import tqdm from PIL import Image RESOLUTION = 256 def load_sidewalks_dataset(args): hf_dataset_identifier = "segments/sidewalk-semantic" ds = datasets.load_dataset(hf_dataset_identifier) ds = ds.shuffle(seed=1) ds = ds["train"].train_test_split(test_size=args.split, seed=args.seed) train_ds = ds["train"] val_ds = ds["test"] return train_ds, val_ds def resize_img( image: tf.Tensor, label: tf.Tensor, resize: int ) -> Tuple[tf.Tensor, tf.Tensor]: image = tf.image.resize(image, (resize, resize)) label = tf.image.resize(label[..., None], (resize, resize)) label = tf.squeeze(label, -1) return image, label def process_image( image: Image, label: Image, resize: int ) -> Tuple[tf.Tensor, tf.Tensor]: image = np.array(image) label = np.array(label) image = tf.convert_to_tensor(image) label = tf.convert_to_tensor(label) if resize: image, label = resize_img(image, label, resize) return image, label def _int64_feature(value): return tf.train.Feature(int64_list=tf.train.Int64List(value=value)) def _float_feature(value): return tf.train.Feature(float_list=tf.train.FloatList(value=value)) def create_tfrecord(image: Image, label: Image, resize: int): image, label = process_image(image, label, resize) image_dims = image.shape label_dims = label.shape image = tf.reshape(image, [-1]) # flatten to 1D array label = tf.reshape(label, [-1]) # flatten to 1D array return tf.train.Example( features=tf.train.Features( feature={ "image": _float_feature(image.numpy()), "image_shape": _int64_feature( [image_dims[0], image_dims[1], image_dims[2]] ), "label": _float_feature(label.numpy()), "label_shape": _int64_feature([label_dims[0], label_dims[1]]), } ) ).SerializeToString() def write_tfrecords(root_dir, dataset, split, batch_size, resize): print(f"Preparing TFRecords for split: {split}.") for step in tqdm.tnrange(int(math.ceil(len(dataset) / batch_size))): temp_ds = dataset[step * batch_size : (step + 1) * batch_size] shard_size = len(temp_ds["pixel_values"]) filename = os.path.join( root_dir, "{}-{:02d}-{}.tfrec".format(split, step, shard_size) ) with tf.io.TFRecordWriter(filename) as out_file: for i in range(shard_size): image = temp_ds["pixel_values"][i] label = temp_ds["label"][i] example = create_tfrecord(image, label, resize) out_file.write(example) print("Wrote file {} containing {} records".format(filename, shard_size)) def main(args): train_ds, val_ds = load_sidewalks_dataset(args) print("Dataset loaded from HF.") if not os.path.exists(args.root_tfrecord_dir): os.makedirs(args.root_tfrecord_dir, exist_ok=True) print(args.resize) write_tfrecords( args.root_tfrecord_dir, train_ds, "train", args.batch_size, args.resize ) write_tfrecords(args.root_tfrecord_dir, val_ds, "val", args.batch_size, args.resize) def parse_args(): parser = argparse.ArgumentParser() parser.add_argument( "--split", help="Train and test split.", default=0.2, type=float ) parser.add_argument( "--seed", help="Seed to be used while performing train-test splits.", default=2022, type=int, ) parser.add_argument( "--root_tfrecord_dir", help="Root directory where the TFRecord shards will be serialized.", default="sidewalks-tfrecords", type=str, ) parser.add_argument( "--batch_size", help="Number of samples to process in a batch before serializing a single TFRecord shard.", default=32, type=int, ) parser.add_argument( "--resize", help="Width and height size the image will be resized to. No resizing will be applied when this isn't set.", type=int, ) args = parser.parse_args() return args if __name__ == "__main__": args = parse_args() main(args)
deep-diver/semantic-segmentation-ml-pipeline
tfrecords/create_tfrecords_pets.py
""" Script to generate TFRecord shards from the Pets dataset as shown in this tutorial: https://keras.io/examples/vision/oxford_pets_image_segmentation/. The recommended way to obtain TFRecord shards is via an Apache Beam Pipeline with an execution runner like Dataflow. Example: https://github.com/GoogleCloudPlatform/practical-ml-vision-book/blob/master/05_create_dataset/jpeg_to_tfrecord.py. Usage: python create_tfrecords_pets.py --batch_size 64 python create_tfrecords_pets.py --resize 128 # without --resize flag, no resizing is applied References: * https://github.com/GoogleCloudPlatform/practical-ml-vision-book/blob/master/05_create_dataset/05_split_tfrecord.ipynb * https://www.tensorflow.org/tutorials/images/segmentation * https://keras.io/examples/vision/oxford_pets_image_segmentation/ """ import argparse import os import random from typing import List, Tuple import numpy as np import tensorflow as tf import tqdm from PIL import Image RESOLUTION = 128 SEED = 2022 def load_paths(args) -> Tuple[List[str], List[str]]: input_img_paths = sorted( [ os.path.join(args.input_dir, fname) for fname in os.listdir(args.input_dir) if fname.endswith(".jpg") ] ) target_img_paths = sorted( [ os.path.join(args.target_dir, fname) for fname in os.listdir(args.target_dir) if fname.endswith(".png") and not fname.startswith(".") ] ) return input_img_paths, target_img_paths def resize_img( image: tf.Tensor, label: tf.Tensor, resize: int ) -> Tuple[tf.Tensor, tf.Tensor]: image = tf.image.resize(image, (resize, resize)) label = tf.image.resize(label[..., None], (resize, resize)) label = tf.squeeze(label, -1) label -= 1 return image, label def process_image( image_path: str, label_path: str, resize: int ) -> Tuple[tf.Tensor, tf.Tensor]: image = Image.open(image_path).convert("RGB") label = Image.open(label_path).convert("L") image = np.array(image) label = np.array(label) image = tf.convert_to_tensor(image) label = tf.convert_to_tensor(label) if resize: image, label = resize_img(image, label, resize) return image, label def split_paths(img_paths: List[str], target_paths: List[str], split: float): val_samples = int(len(img_paths) * split) random.Random(SEED).shuffle(img_paths) random.Random(SEED).shuffle(target_paths) train_img_paths = img_paths[:-val_samples] train_target_paths = target_paths[:-val_samples] val_img_paths = img_paths[-val_samples:] val_target_paths = target_paths[-val_samples:] return train_img_paths, train_target_paths, val_img_paths, val_target_paths def prepare_tf_dataset( img_paths: List[str], target_paths: List[str], batch_size: int ) -> tf.data.Dataset: tf_dataset = tf.data.Dataset.from_tensor_slices((img_paths, target_paths)) return tf_dataset.batch(batch_size) def get_tf_datasets(args) -> Tuple[tf.data.Dataset, tf.data.Dataset]: input_img_paths, target_img_paths = load_paths(args) train_img_paths, train_target_paths, val_img_paths, val_target_paths = split_paths( input_img_paths, target_img_paths, args.split ) train_ds = prepare_tf_dataset(train_img_paths, train_target_paths, args.batch_size) val_ds = prepare_tf_dataset(val_img_paths, val_target_paths, args.batch_size) return train_ds, val_ds def _int64_feature(value): return tf.train.Feature(int64_list=tf.train.Int64List(value=value)) def _float_feature(value): return tf.train.Feature(float_list=tf.train.FloatList(value=value)) def create_tfrecord(image: Image, label: Image, resize: int): image, label = process_image(image, label, resize) image_dims = image.shape label_dims = label.shape image = tf.reshape(image, [-1]) # flatten to 1D array label = tf.reshape(label, [-1]) # flatten to 1D array return tf.train.Example( features=tf.train.Features( feature={ "image": _float_feature(image.numpy()), "image_shape": _int64_feature( [image_dims[0], image_dims[1], image_dims[2]] ), "label": _float_feature(label.numpy()), "label_shape": _int64_feature([label_dims[0], label_dims[1]]), } ) ).SerializeToString() def write_tfrecords(root_dir: str, dataset: tf.data.Dataset, split: str, resize: int): print(f"Preparing TFRecords for split: {split}.") for shard, (image_paths, label_paths) in enumerate(tqdm.tqdm(dataset)): shard_size = image_paths.numpy().shape[0] filename = os.path.join( root_dir, "{}-{:02d}-{}.tfrec".format(split, shard, shard_size) ) with tf.io.TFRecordWriter(filename) as out_file: for i in range(shard_size): img_path = image_paths[i] label_path = label_paths[i] example = create_tfrecord(img_path.numpy(), label_path.numpy(), resize) out_file.write(example) print("Wrote file {} containing {} records".format(filename, shard_size)) def main(args): train_ds, val_ds = get_tf_datasets(args) print("TensorFlow datasets loaded.") if not os.path.exists(args.root_tfrecord_dir): os.makedirs(args.root_tfrecord_dir, exist_ok=True) write_tfrecords(args.root_tfrecord_dir, train_ds, "train", args.resize) write_tfrecords(args.root_tfrecord_dir, val_ds, "val", args.resize) def parse_args(): parser = argparse.ArgumentParser() parser.add_argument("--split", help="Train and test split.", default=0.2, type=float) parser.add_argument( "--input_dir", help="Path to the directory containing all images.", default="images/", type=str, ) parser.add_argument( "--target_dir", help="Path to the directory containing all targets.", default="annotations/trimaps/", type=str, ) parser.add_argument( "--root_tfrecord_dir", help="Root directory where the TFRecord shards will be serialized.", default="pets-tfrecords", type=str, ) parser.add_argument( "--batch_size", help="Number of samples to process in a batch before serializing a single TFRecord shard.", default=32, type=int, ) parser.add_argument( "--resize", help="Width and height size the image will be resized to. No resizing will be applied when this isn't set.", type=int, ) args = parser.parse_args() return args if __name__ == "__main__": args = parse_args() main(args)
deep-diver/semantic-segmentation-ml-pipeline
tfrecords/create_tfrecords_str.py
""" Script to generate TFRecord shards from the Sidewalks dataset as shown in this blog post: https://huggingface.co/blog/fine-tune-segformer. The recommended way to obtain TFRecord shards is via an Apache Beam Pipeline with an execution runner like Dataflow. Example: https://github.com/GoogleCloudPlatform/practical-ml-vision-book/blob/master/05_create_dataset/jpeg_to_tfrecord.py. Usage: python create_tfrecords --batch_size 16 References: * https://github.com/sayakpaul/TF-2.0-Hacks/blob/master/Cats_vs_Dogs_TFRecords.ipynb * https://www.tensorflow.org/tutorials/images/segmentation """ import argparse import math import os from typing import Tuple import datasets import numpy as np import tensorflow as tf import tqdm from PIL import Image RESOLUTION = 256 def load_sidewalks_dataset(args): hf_dataset_identifier = "segments/sidewalk-semantic" ds = datasets.load_dataset(hf_dataset_identifier) ds = ds.shuffle(seed=1) ds = ds["train"].train_test_split(test_size=args.split, seed=args.seed) train_ds = ds["train"] val_ds = ds["test"] return train_ds, val_ds def resize_img(image: tf.Tensor, label: tf.Tensor) -> Tuple[tf.Tensor, tf.Tensor]: image = tf.image.resize(image, (RESOLUTION, RESOLUTION)) label = tf.image.resize(label[..., None], (RESOLUTION, RESOLUTION)) label = tf.squeeze(label, -1) return image, label def process_image(image: Image, label: Image) -> Tuple[tf.Tensor, tf.Tensor]: image = np.array(image) label = np.array(label) image = tf.convert_to_tensor(image) label = tf.convert_to_tensor(label) image, label = resize_img(image, label) image, label = normalize_img(image, label) return image, label def _bytestring_feature(value): return tf.train.Feature(bytes_list=tf.train.BytesList(value=value)) def create_tfrecord(image: Image, label: Image): image, label = process_image(image, label) image = tf.io.serialize_tensor(image) label = tf.io.serialize_tensor(label) return tf.train.Example( features=tf.train.Features( feature={ "image": _bytestring_feature([image.numpy()]), "label": _bytestring_feature([label.numpy()]), } ) ).SerializeToString() def write_tfrecords(root_dir, dataset, split, batch_size): print(f"Preparing TFRecords for split: {split}.") for step in tqdm.tnrange(int(math.ceil(len(dataset) / batch_size))): temp_ds = dataset[step * batch_size : (step + 1) * batch_size] shard_size = len(temp_ds["pixel_values"]) filename = os.path.join( root_dir, "{}-{:02d}-{}.tfrec".format(split, step, shard_size) ) with tf.io.TFRecordWriter(filename) as out_file: for i in range(shard_size): image = temp_ds["pixel_values"][i] label = temp_ds["label"][i] example = create_tfrecord(image, label) out_file.write(example) print("Wrote file {} containing {} records".format(filename, shard_size)) def main(args): train_ds, val_ds = load_sidewalks_dataset(args) print("Dataset loaded from HF.") if not os.path.exists(args.root_tfrecord_dir): os.makedirs(args.root_tfrecord_dir, exist_ok=True) write_tfrecords(args.root_tfrecord_dir, train_ds, "train", args.batch_size) write_tfrecords(args.root_tfrecord_dir, val_ds, "val", args.batch_size) def parse_args(): parser = argparse.ArgumentParser() parser.add_argument( "--split", help="Train and test split.", default=0.2, type=float ) parser.add_argument( "--seed", help="Seed to be used while performing train-test splits.", default=2022, type=int, ) parser.add_argument( "--root_tfrecord_dir", help="Root directory where the TFRecord shards will be serialized.", default="sidewalks-tfrecords", type=str, ) parser.add_argument( "--batch_size", help="Number of samples to process in a batch before serializing a single TFRecord shard.", default=32, type=int, ) args = parser.parse_args() return args if __name__ == "__main__": args = parse_args() main(args)
deep-diver/semantic-segmentation-ml-pipeline
training_pipeline/__init__.py
# Copyright 2020 Google LLC. All Rights Reserved. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License.
deep-diver/semantic-segmentation-ml-pipeline
training_pipeline/data_validation.ipynb
# import required libs import glob import os import tensorflow as tf import tensorflow_data_validation as tfdv print('TF version: {}'.format(tf.version.VERSION)) print('TFDV version: {}'.format(tfdv.version.__version__))# Read artifact information from metadata store. import beam_dag_runner from tfx.orchestration import metadata from tfx.types import standard_artifacts metadata_connection_config = metadata.sqlite_metadata_connection_config( beam_dag_runner.METADATA_PATH) with metadata.Metadata(metadata_connection_config) as store: stats_artifacts = store.get_artifacts_by_type(standard_artifacts.ExampleStatistics.TYPE_NAME) schema_artifacts = store.get_artifacts_by_type(standard_artifacts.Schema.TYPE_NAME) anomalies_artifacts = store.get_artifacts_by_type(standard_artifacts.ExampleAnomalies.TYPE_NAME)# configure output paths # Exact paths to output artifacts can also be found on KFP Web UI if you are using kubeflow. stats_path = stats_artifacts[-1].uri train_stats_file = os.path.join(stats_path, 'train', 'stats_tfrecord') eval_stats_file = os.path.join(stats_path, 'eval', 'stats_tfrecord') print("Train stats file:{}, Eval stats file:{}".format( train_stats_file, eval_stats_file)) schema_file = os.path.join(schema_artifacts[-1].uri, 'schema.pbtxt') print("Generated schame file:{}".format(schema_file)) anomalies_file = os.path.join(anomalies_artifacts[-1].uri, 'anomalies.pbtxt') print("Generated anomalies file:{}".format(anomalies_file))# load generated statistics from StatisticsGen train_stats = tfdv.load_statistics(train_stats_file) eval_stats = tfdv.load_statistics(eval_stats_file) tfdv.visualize_statistics(lhs_statistics=eval_stats, rhs_statistics=train_stats, lhs_name='EVAL_DATASET', rhs_name='TRAIN_DATASET')# load generated schema from SchemaGen schema = tfdv.load_schema_text(schema_file) tfdv.display_schema(schema=schema)# load data vaildation result from ExampleValidator anomalies = tfdv.load_anomalies_text(anomalies_file) tfdv.display_anomalies(anomalies)