Plonk / data /webdataset.py
nicolas-dufour's picture
squash: merge all unpushed commits
c4c7cee
raw
history blame
13.8 kB
import glob
import json
import logging
import os
import random
from collections import OrderedDict
from multiprocessing import Value
from pathlib import Path
import braceexpand
import numpy as np
import pandas as pd
import torch
import webdataset as wds
from lightning_fabric.utilities.rank_zero import _get_rank
from PIL import Image
from torch.utils.data import Dataset, get_worker_info
from tqdm import tqdm
from webdataset.tariterators import (
base_plus_ext,
tar_file_expander,
url_opener,
valid_sample,
)
from functools import partial
import math
class GPSWebdataset(wds.DataPipeline):
def __init__(
self,
root,
image_transforms=None,
distributed=True,
train=True,
epoch=0,
seed=3407,
embedding_name=None,
return_image=True,
shard_shuffle_size=2000,
shard_shuffle_initial=500,
sample_shuffle_size=5000,
sample_shuffle_initial=1000,
metadata_attributes=[],
):
self.image_transforms = image_transforms
dataset_tar_files = []
# Get a list of all tar files in the directory
if " " in root:
root = root.split(" ")
print(f"Using multiple dataset[s: {root}")
if isinstance(root, str):
tar_files = [f for f in os.listdir(root) if f.endswith(".tar")]
# Sort the list of tar files
tar_files.sort()
first_tar_file = tar_files[0].split(".")[0]
last_tar_file = tar_files[-1].split(".")[0]
for tar_file in tar_files:
dataset_tar_files.append(f"{root}/{tar_file}")
dataset_pattern = f"{root}/{{{first_tar_file}..{last_tar_file}}}.tar"
self.num_samples, _ = get_dataset_size(dataset_pattern)
elif isinstance(root, list):
num_samples = 0
for r in root:
tar_files = [f for f in os.listdir(r) if f.endswith(".tar")]
tar_files.sort()
first_tar_file = tar_files[0].split(".")[0]
last_tar_file = tar_files[-1].split(".")[0]
for tar_file in tar_files:
dataset_tar_files.append(f"{r}/{tar_file}")
num_samples += get_dataset_size(
f"{r}/{{{first_tar_file}..{last_tar_file}}}.tar"
)[0]
self.num_samples = num_samples
else:
raise ValueError(
f"root must be a string or list of strings. Got {type(root)}"
)
rank = _get_rank()
self.shared_epoch = SharedEpoch(epoch)
pipeline = [wds.SimpleShardList(dataset_tar_files)]
if distributed:
if train:
pipeline.extend(
[
detshuffle2(
bufsize=shard_shuffle_size,
initial=shard_shuffle_initial,
seed=seed,
epoch=self.shared_epoch,
),
wds.split_by_node,
wds.split_by_worker,
tarfile_to_samples_nothrow,
wds.shuffle(
bufsize=sample_shuffle_size,
initial=sample_shuffle_initial,
),
]
)
else:
pipeline.extend(
[wds.split_by_node, wds.split_by_worker, tarfile_to_samples_nothrow]
)
else:
if train:
pipeline.extend(
[
wds.shuffle(
bufsize=shard_shuffle_size,
initial=sample_shuffle_initial,
),
wds.split_by_worker,
tarfile_to_samples_nothrow,
wds.shuffle(
bufsize=sample_shuffle_size,
initial=sample_shuffle_initial,
),
]
)
else:
pipeline.extend([wds.split_by_worker, tarfile_to_samples_nothrow])
outputs_transforms = OrderedDict()
outputs_rename = OrderedDict()
if return_image:
outputs_rename["img.jpg"] = "jpg;png;webp;jpeg"
outputs_transforms["img.jpg"] = (
self.image_transforms
if self.image_transforms is not None
else lambda x: x
)
if embedding_name is not None:
outputs_rename[f"emb.npy"] = f"{embedding_name}.npy"
outputs_transforms[f"emb.npy"] = lambda x: torch.from_numpy(x)
if metadata_attributes != []:
for attr in metadata_attributes:
outputs_rename[f"{attr}.json"] = f"json"
outputs_transforms[f"{attr}.json"] = partial(get_attr, attr=attr)
outputs_rename["gps"] = "json"
outputs_transforms["gps"] = get_gps
pipeline.extend(
[
wds.rename(**outputs_rename),
filter_dict_keys(*outputs_rename.keys(), handler=log_and_continue),
]
)
if return_image:
pipeline.append(wds.decode("pilrgb", handler=log_and_continue))
else:
pipeline.append(wds.decode(handler=log_and_continue))
pipeline.extend(
[
wds.map_dict(**outputs_transforms, handler=log_and_continue),
wds.rename(
**{k.split(".")[0]: k for k in outputs_transforms.keys()},
),
]
)
super().__init__(*pipeline)
def __len__(self):
return self.num_samples
def normalize_gps(lat, lon):
"""Used to put all lat lon inside ±90 and ±180."""
lat = (lat + 90) % 360 - 90
if lat > 90:
lat = 180 - lat
lon += 180
lon = (lon + 180) % 360 - 180
return lat, lon
def get_attr(metadata, attr):
# datapoint = json.loads(metadata)
attr_value = metadata[attr]
if isinstance(attr_value, float) and math.isnan(attr_value):
return "NaN"
else:
return attr_value
def get_gps(metadata):
datapoint = json.loads(metadata)
lat, lon = normalize_gps(
float(datapoint["latitude"]), float(datapoint["longitude"])
)
gps = torch.tensor([np.radians(lat), np.radians(lon)], dtype=torch.float)
return gps
def get_dataset_size(shards):
shards_list, _ = expand_urls(shards)
dir_path = os.path.dirname(shards_list[0])
sizes_filename = os.path.join(dir_path, "sizes.json")
if os.path.exists(sizes_filename):
sizes = json.load(open(sizes_filename, "r"))
total_size = sum([int(sizes[os.path.basename(shard)]) for shard in shards_list])
else:
total_size = 0 # num samples undefined
sizes = {}
for shard in tqdm(shards_list):
dataset = wds.WebDataset(shard)
num_samples = sum(1 for _ in dataset)
total_size += num_samples
sizes[os.path.basename(shard)] = num_samples
print(f"Total number of samples: {total_size}")
with open(sizes_filename, "w") as f:
json.dump(sizes, f)
num_shards = len(shards_list)
return total_size, num_shards
def expand_urls(urls, weights=None):
if weights is None:
expanded_urls = wds.shardlists.expand_urls(urls)
return expanded_urls, None
if isinstance(urls, str):
urllist = urls.split("::")
weights = weights.split("::")
assert len(weights) == len(
urllist
), f"Expected the number of data components ({len(urllist)}) and weights({len(weights)}) to match."
weights = [float(weight) for weight in weights]
all_urls, all_weights = [], []
for url, weight in zip(urllist, weights):
expanded_url = list(braceexpand.braceexpand(url))
expanded_weights = [weight for _ in expanded_url]
all_urls.extend(expanded_url)
all_weights.extend(expanded_weights)
return all_urls, all_weights
else:
all_urls = list(urls)
return all_urls, weights
class SharedEpoch:
def __init__(self, epoch: int = 0):
self.shared_epoch = Value("i", epoch)
def set_value(self, epoch):
self.shared_epoch.value = epoch
def get_value(self):
return self.shared_epoch.value
# _SHARD_SHUFFLE_SIZE = 256
# _SHARD_SHUFFLE_INITIAL = 128
# _SAMPLE_SHUFFLE_SIZE = 5000
# _SAMPLE_SHUFFLE_INITIAL = 1000
class detshuffle2(wds.PipelineStage):
def __init__(
self,
bufsize=1000,
initial=100,
seed=0,
epoch=-1,
):
self.bufsize = bufsize
self.initial = initial
self.seed = seed
self.epoch = epoch
def run(self, src):
if isinstance(self.epoch, SharedEpoch):
epoch = self.epoch.get_value()
else:
# NOTE: this is epoch tracking is problematic in a multiprocess (dataloader workers or train)
# situation as different workers may wrap at different times (or not at all).
self.epoch += 1
epoch = self.epoch
rng = random.Random()
if self.seed < 0:
# If seed is negative, we use the worker's seed, this will be different across all nodes/workers
seed = pytorch_worker_seed(epoch)
else:
# This seed to be deterministic AND the same across all nodes/workers in each epoch
seed = self.seed + epoch
rng.seed(seed)
return wds.filters._shuffle(src, self.bufsize, self.initial, rng)
def pytorch_worker_seed(increment=0):
"""get dataloader worker seed from pytorch"""
worker_info = get_worker_info()
if worker_info is not None:
# favour using the seed already created for pytorch dataloader workers if it exists
seed = worker_info.seed
if increment:
# space out seed increments so they can't overlap across workers in different iterations
seed += increment * max(1, worker_info.num_workers)
return seed
# fallback to wds rank based seed
return wds.utils.pytorch_worker_seed()
def log_and_continue(exn):
"""Call in an exception handler to ignore any exception, issue a warning, and continue."""
logging.warning(f"Handling webdataset error ({repr(exn)}). Ignoring.")
return True
def group_by_keys_nothrow(
data, keys=base_plus_ext, lcase=True, suffixes=None, handler=None
):
"""Return function over iterator that groups key, value pairs into samples.
:param keys: function that splits the key into key and extension (base_plus_ext)
:param lcase: convert suffixes to lower case (Default value = True)
"""
current_sample = None
for filesample in data:
assert isinstance(filesample, dict)
fname, value = filesample["fname"], filesample["data"]
prefix, suffix = keys(fname)
if prefix is None:
continue
if lcase:
suffix = suffix.lower()
# FIXME webdataset version throws if suffix in current_sample, but we have a potential for
# this happening in the current LAION400m dataset if a tar ends with same prefix as the next
# begins, rare, but can happen since prefix aren't unique across tar files in that dataset
if (
current_sample is None
or prefix != current_sample["__key__"]
or suffix in current_sample
):
if valid_sample(current_sample):
yield current_sample
current_sample = dict(__key__=prefix, __url__=filesample["__url__"])
if suffixes is None or suffix in suffixes:
current_sample[suffix] = value
if valid_sample(current_sample):
yield current_sample
def tarfile_to_samples_nothrow(src, handler=log_and_continue):
# NOTE this is a re-impl of the webdataset impl with group_by_keys that doesn't throw
streams = url_opener(src, handler=handler)
files = tar_file_expander(streams, handler=handler)
samples = group_by_keys_nothrow(files, handler=handler)
return samples
def filter_no_caption_or_no_image(sample):
has_caption = "txt" in sample
has_image = (
"png" in sample or "jpg" in sample or "jpeg" in sample or "webp" in sample
)
return has_caption and has_image
def filter_metadata(sample, min_image_size, min_clip_score):
metadata = json.loads(sample["json"])
width = metadata["width"]
height = metadata["height"]
clip_score = metadata["clip_score"] / 100
return (
width >= min_image_size
and height >= min_image_size
and clip_score >= min_clip_score
)
def _filter_dict_keys(
data,
*args,
handler=wds.reraise_exception,
missing_is_error=True,
none_is_error=None,
):
"""Convert dict samples to tuples."""
if none_is_error is None:
none_is_error = missing_is_error
if len(args) == 1 and isinstance(args[0], str) and " " in args[0]:
args = args[0].split()
for sample in data:
try:
result = {
f: wds.getfirst(sample, f, missing_is_error=missing_is_error)
for f in args
}
print
if none_is_error and any(x is None for x in result):
raise ValueError(f"to_tuple {args} got {sample.keys()}")
yield result
except Exception as exn:
if handler(exn):
continue
else:
break
filter_dict_keys = wds.pipelinefilter(_filter_dict_keys)