AquaLearn / app.py
Sanshruth's picture
Upload 2 files
83d8f3b verified
raw
history blame
15.7 kB
import streamlit as st
import h2o
from h2o.automl import H2OAutoML
import pandas as pd
import os
import numpy as np
from sklearn.metrics import mean_squared_error
import matplotlib.pyplot as plt
import shutil
import zipfile
import io
import tempfile
import zipfile
# Set page config at the very beginning
st.set_page_config(page_title="AquaLearn", layout="wide")
# Initialize the H2O server
h2o.init()
def rename_columns_alphabetically(df):
new_columns = [chr(65 + i) for i in range(len(df.columns))]
return df.rename(columns=dict(zip(df.columns, new_columns)))
def sanitize_column_name(name):
# Replace non-alphanumeric characters with underscores
sanitized = ''.join(c if c.isalnum() else '_' for c in name)
# Ensure the name starts with a letter or underscore
if not sanitized[0].isalpha() and sanitized[0] != '_':
sanitized = 'f_' + sanitized
return sanitized
# Create a directory for saving models
if not os.path.exists("saved_models"):
os.makedirs("saved_models")
def load_data():
st.title("Aqua Learn")
uploaded_file = st.file_uploader("Choose a CSV file", type="csv")
if uploaded_file is not None:
train = pd.read_csv(uploaded_file)
st.write(train.head())
return h2o.H2OFrame(train)
return None
def select_problem_type():
return st.selectbox("Select Problem Type:", ['Classification', 'Regression'])
def select_target_column(train_h2o):
return st.selectbox("Select Target Column:", train_h2o.columns)
def prepare_features(train_h2o, y, problem_type):
x = train_h2o.columns
x.remove(y)
if problem_type == 'Classification':
train_h2o[y] = train_h2o[y].asfactor()
# Rename columns
new_columns = [chr(65 + i) for i in range(len(train_h2o.columns))]
train_h2o.columns = new_columns
y = new_columns[-1] # Assume the target is the last column
x = new_columns[:-1]
return x, y, train_h2o
def select_algorithms():
algorithm_options = ['DeepLearning', 'GLM', 'GBM', 'DRF', 'XGBoost']
return st.multiselect("Select Algorithms:", algorithm_options)
def set_automl_parameters():
max_models = st.number_input("Max Models:", value=20, min_value=1)
max_runtime = st.number_input("Max Runtime (seconds):", value=600, min_value=1)
return max_models, max_runtime
def run_automl(x, y, train, problem_type, selected_algos, max_models, max_runtime):
aml = H2OAutoML(max_models=max_models,
seed=1,
max_runtime_secs=max_runtime,
sort_metric="AUC" if problem_type == 'Classification' else "RMSE",
include_algos=selected_algos)
aml.train(x=x, y=y, training_frame=train)
return aml
def display_results(aml, test):
st.subheader("AutoML Leaderboard")
st.write(aml.leaderboard.as_data_frame())
st.subheader("Best Model Performance")
best_model = aml.leader
perf = best_model.model_performance(test)
st.write(perf)
def save_and_evaluate_models(aml, test, y, problem_type):
if st.button("Save Models and Calculate Performance"):
model_performances = []
for model_id in aml.leaderboard['model_id'].as_data_frame().values:
model = h2o.get_model(model_id[0])
# model_path = os.path.join("saved_models", f"{model_id[0]}")
# h2o.save_model(model=model, path=model_path, force=True)
# st.session_state.saved_models.append((model_id[0], model_path))
preds = model.predict(test)
actual = test[y].as_data_frame().values.flatten()
predicted = preds.as_data_frame()['predict'].values.flatten()
if problem_type == 'Classification':
performance = (actual == predicted).mean()
metric_name = 'accuracy'
else:
performance = np.sqrt(mean_squared_error(actual, predicted))
metric_name = 'rmse'
model_performances.append({'model_id': model_id[0], metric_name: performance})
performance_df = pd.DataFrame(model_performances)
st.write(performance_df)
# Create and display the bar plot
st.subheader("Model Performance Visualization")
fig, ax = plt.subplots(figsize=(10, 6))
performance_df.sort_values(by=metric_name, ascending=False, inplace=True)
ax.barh(performance_df['model_id'], performance_df[metric_name], color='skyblue')
ax.set_xlabel(metric_name.capitalize())
ax.set_ylabel('Model ID')
ax.set_title(f'Model {metric_name.capitalize()} from H2O AutoML')
ax.grid(axis='x')
st.pyplot(fig)
def download_model():
st.subheader("Download Model")
if 'saved_models' in st.session_state and st.session_state.saved_models:
model_to_download = st.selectbox("Select Model to Download:",
[model[0] for model in st.session_state.saved_models])
if st.button("Download Selected Model"):
model_path = next(model[1] for model in st.session_state.saved_models if model[0] == model_to_download)
if os.path.isdir(model_path):
# If it's a directory, create a zip file
zip_buffer = io.BytesIO()
with zipfile.ZipFile(zip_buffer, 'w', zipfile.ZIP_DEFLATED) as zip_file:
for root, _, files in os.walk(model_path):
for file in files:
zip_file.write(os.path.join(root, file),
os.path.relpath(os.path.join(root, file), model_path))
zip_buffer.seek(0)
st.download_button(
label="Click to Download",
data=zip_buffer,
file_name=f"{model_to_download}.zip",
mime="application/zip"
)
else:
# If it's already a file, offer it for download
with open(model_path, "rb") as file:
st.download_button(
label="Click to Download",
data=file,
file_name=f"{model_to_download}.zip",
mime="application/zip"
)
else:
st.write("No models available for download. Please train and save models first.")
def further_training(aml, x, y, train, problem_type):
st.subheader("Further Training")
leaderboard_df = aml.leaderboard.as_data_frame()
model_to_train = st.selectbox("Select Model for Training:", leaderboard_df['model_id'].tolist())
training_time = st.number_input("Training Time (seconds):", value=60, min_value=1)
if st.button("Train Model"):
model = h2o.get_model(model_to_train)
with st.spinner(f"Training model: {model_to_train} for {training_time} seconds..."):
if isinstance(model, h2o.estimators.stackedensemble.H2OStackedEnsembleEstimator):
aml = H2OAutoML(max_runtime_secs=training_time, seed=1, sort_metric="AUC" if problem_type == 'Classification' else "RMSE")
aml.train(x=x, y=y, training_frame=train)
model = aml.leader
else:
model.train(x=x, y=y, training_frame=train, max_runtime_secs=training_time)
perf = model.model_performance(train)
st.write("Model performance after training:")
st.write(perf)
# Create a temporary directory to save the model
temp_dir = os.path.join("saved_models", "temp")
os.makedirs(temp_dir, exist_ok=True)
model_path = os.path.join(temp_dir, f"{model.model_id}")
h2o.save_model(model=model, path=model_path, force=True)
# Create a zip file of the model
zip_buffer = io.BytesIO()
with zipfile.ZipFile(zip_buffer, 'w', zipfile.ZIP_DEFLATED) as zip_file:
for root, _, files in os.walk(model_path):
for file in files:
zip_file.write(os.path.join(root, file),
os.path.relpath(os.path.join(root, file), model_path))
zip_buffer.seek(0)
st.download_button(
label="Download Retrained Model",
data=zip_buffer,
file_name=f"{model.model_id}.zip",
mime="application/zip"
)
# Clean up the temporary directory
shutil.rmtree(temp_dir)
st.success(f"Retrained model ready for download: {model.model_id}")
def make_prediction():
st.subheader("Make Prediction")
uploaded_zip = st.file_uploader("Upload a zip file containing the model", type="zip")
if uploaded_zip is not None:
with tempfile.TemporaryDirectory() as tmpdirname:
zip_path = os.path.join(tmpdirname, "model.zip")
with open(zip_path, "wb") as f:
f.write(uploaded_zip.getbuffer())
with zipfile.ZipFile(zip_path, 'r') as zip_ref:
zip_ref.extractall(tmpdirname)
extracted_files = os.listdir(tmpdirname)
if len(extracted_files) == 0:
st.error("The uploaded zip file is empty.")
return
model_file = next((f for f in extracted_files if f != "model.zip"), None)
if model_file is None:
st.error("No model file found in the uploaded zip.")
return
model_path = os.path.join(tmpdirname, model_file)
try:
model_for_prediction = h2o.load_model(model_path)
except Exception as e:
st.error(f"Error loading the model: {str(e)}")
st.error("Please ensure you're uploading a valid H2O model file.")
return
# Ask user to input feature names
feature_names_input = st.text_input("Enter feature names, separated by commas:")
original_feature_names = [name.strip() for name in feature_names_input.split(',') if name.strip()]
if not original_feature_names:
st.error("Please enter at least one feature name.")
return
# Create a mapping from original names to A, B, C, etc.
feature_mapping = {name: chr(65 + i) for i, name in enumerate(original_feature_names)}
reverse_mapping = {v: k for k, v in feature_mapping.items()}
prediction_type = st.radio("Choose prediction type:", ["Upload CSV", "Single Entry"])
if prediction_type == "Upload CSV":
uploaded_csv = st.file_uploader("Upload a CSV file for prediction", type="csv")
if uploaded_csv is not None:
prediction_data = pd.read_csv(uploaded_csv)
# Rename columns to A, B, C, etc.
prediction_data = prediction_data.rename(columns=feature_mapping)
prediction_h2o = h2o.H2OFrame(prediction_data)
try:
predictions = model_for_prediction.predict(prediction_h2o)
predictions_df = predictions.as_data_frame()
# Combine original data with predictions
result_df = pd.concat([prediction_data, predictions_df], axis=1)
# Rename columns back to original names for display
result_df = result_df.rename(columns=reverse_mapping)
st.write("Predictions (showing first 10 rows):")
st.write(result_df.head(10))
# Option to download the full results
csv = result_df.to_csv(index=False)
st.download_button(
label="Download full results as CSV",
data=csv,
file_name="predictions_results.csv",
mime="text/csv"
)
except Exception as e:
st.error(f"Error making predictions: {str(e)}")
st.error("Please ensure your CSV file matches the model's expected input format.")
else: # Single Entry
sample_input = {}
for original_name, coded_name in feature_mapping.items():
value = st.text_input(f"Enter {original_name} ({coded_name}):")
try:
sample_input[coded_name] = [float(value)]
except ValueError:
sample_input[coded_name] = [value]
if st.button("Predict"):
sample_h2o = h2o.H2OFrame(sample_input)
try:
predictions = model_for_prediction.predict(sample_h2o)
prediction_value = predictions['predict'][0,0]
st.write(f"Predicted value: {prediction_value}")
except Exception as e:
st.error(f"Error making prediction: {str(e)}")
st.error("Please ensure you've entered valid input values.")
else:
st.write("Please upload a zip file containing the model to make predictions.")
def main():
train_h2o = load_data()
if train_h2o is not None:
problem_type = select_problem_type()
target_column = select_target_column(train_h2o)
if st.button("Set Target and Continue"):
x, target_column, train_h2o = prepare_features(train_h2o, target_column, problem_type)
st.session_state.features_prepared = True
st.session_state.x = x
st.session_state.target_column = target_column
st.session_state.train_h2o = train_h2o
st.session_state.problem_type = problem_type
if 'features_prepared' in st.session_state and st.session_state.features_prepared:
st.write(f"Target Column: {st.session_state.target_column}")
st.write(f"Feature Columns: {st.session_state.x}")
train, test = st.session_state.train_h2o.split_frame(ratios=[0.8])
selected_algos = select_algorithms()
max_models, max_runtime = set_automl_parameters()
if st.button("Start AutoML"):
if not selected_algos:
st.error("Please select at least one algorithm.")
else:
with st.spinner("Running AutoML..."):
aml = run_automl(st.session_state.x, st.session_state.target_column, train,
st.session_state.problem_type, selected_algos, max_models, max_runtime)
st.success("AutoML training completed.")
st.session_state.aml = aml
st.session_state.test = test
if 'aml' in st.session_state:
display_results(st.session_state.aml, st.session_state.test)
save_and_evaluate_models(st.session_state.aml, st.session_state.test, st.session_state.target_column, st.session_state.problem_type)
download_model()
further_training(st.session_state.aml, st.session_state.x, st.session_state.target_column, train, st.session_state.problem_type)
make_prediction() # Call make_prediction without arguments
if __name__ == "__main__":
if 'features_prepared' not in st.session_state:
st.session_state.features_prepared = False
if 'saved_models' not in st.session_state:
st.session_state.saved_models = []
main()
# Clean up saved models when the script ends
shutil.rmtree("saved_models", ignore_errors=True)