Skip to content

Add option to cache transformed data from processors and skip pipeline entirely #783

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
jhnwu3 merged 4 commits into from
Jan 14, 2026
Merged

Add option to cache transformed data from processors and skip pipeline entirely #783

Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
49ea92d
skip proc transformation if it already exists
Logiquo Jan 14, 2026
7c1ec3a
Fix sort_key
Logiquo Jan 14, 2026
616e87b
Fix test
Logiquo Jan 14, 2026
b6a89c8
remove hex for backward compitability
Logiquo Jan 14, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
92 changes: 61 additions & 31 deletions pyhealth/datasets/base_dataset.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -849,44 +849,74 @@ def set_task(
cache_dir = Path(cache_dir)
cache_dir.mkdir(parents=True, exist_ok=True)

proc_params = json.dumps(
{
"input_schema": task.input_schema,
"output_schema": task.output_schema,
"input_processors": (
{
f"{k}_{v.__class__.__name__}": vars(v)
for k, v in input_processors.items()
}
if input_processors
else None
),
"output_processors": (
{
f"{k}_{v.__class__.__name__}": vars(v)
for k, v in output_processors.items()
}
if output_processors
else None
),
},
sort_keys=True,
default=str
)

task_df_path = Path(cache_dir) / "task_df.ld"
samples_path = Path(cache_dir) / f"samples_{uuid.uuid4()}.ld"
samples_path = Path(cache_dir) / f"samples_{uuid.uuid5(uuid.NAMESPACE_DNS, proc_params)}.ld"

task_df_path.mkdir(parents=True, exist_ok=True)
samples_path.mkdir(parents=True, exist_ok=True)

if not (samples_path / "index.json").exists():
# Check if index.json exists to verify cache integrity, this
# is the standard file for litdata.StreamingDataset
if not (task_df_path / "index.json").exists():
self._task_transform(
task,
task_df_path,
num_workers,
)
else:
logger.info(f"Found cached task dataframe at {task_df_path}, skipping task transformation.")

# Check if index.json exists to verify cache integrity, this
# is the standard file for litdata.StreamingDataset
if not (task_df_path / "index.json").exists():
self._task_transform(
task,
# Build processors and fit on the dataset
logger.info(f"Fitting processors on the dataset...")
dataset = litdata.StreamingDataset(
str(task_df_path),
transform=lambda x: pickle.loads(x["sample"]),
)
builder = SampleBuilder(
input_schema=task.input_schema, # type: ignore
output_schema=task.output_schema, # type: ignore
input_processors=input_processors,
output_processors=output_processors,
)
builder.fit(dataset)
builder.save(str(samples_path / "schema.pkl"))

# Apply processors and save final samples to cache_dir
logger.info(f"Processing samples and saving to {samples_path}...")
self._proc_transform(
task_df_path,
samples_path,
num_workers,
)

# Build processors and fit on the dataset
logger.info(f"Fitting processors on the dataset...")
dataset = litdata.StreamingDataset(
str(task_df_path),
transform=lambda x: pickle.loads(x["sample"]),
)
builder = SampleBuilder(
input_schema=task.input_schema, # type: ignore
output_schema=task.output_schema, # type: ignore
input_processors=input_processors,
output_processors=output_processors,
)
builder.fit(dataset)
builder.save(str(samples_path / "schema.pkl"))

# Apply processors and save final samples to cache_dir
logger.info(f"Processing samples and saving to {samples_path}...")
self._proc_transform(
task_df_path,
samples_path,
num_workers,
)
logger.info(f"Cached processed samples to {samples_path}")
logger.info(f"Cached processed samples to {samples_path}")
else:
logger.info(f"Found cached processed samples at {samples_path}, skipping processing.")

return SampleDataset(
path=str(samples_path),
Expand All @@ -902,4 +932,4 @@ def _main_guard(self, func_name: str):
f"{func_name} method accessed from a non-main process. This may lead to unexpected behavior.\n"
+ "Consider use __name__ == '__main__' guard when using multiprocessing."
)
exit(1)
exit(1)