pyg-team/pytorch_geometric

Support filesystem abstraction with fsspec or epath.

Open

#8,336 opened on Nov 7, 2023

View on GitHub
 (9 comments) (1 reaction) (1 assignee)Python (3,514 forks)batch import
datasetfeaturehelp wanted

Repository metrics

Stars
 (19,985 stars)
PR merge metrics
 (Avg merge 35d 1h) (14 merged PRs in 30d)

Description

🛠 Proposed Refactor

I propose that we replace usage of core python I/O e.g., open, exists, is_file, makedirs, with equivalent filesystem abstraction that allows dispatch to appropriate filesystem backends. This would allow users of your library to easily work with remote filesystems as they work with local filesystems.

E.g.,

from torch_geometric.datasets import Planetoid
from torch_geometric.transforms import NormalizeFeatures
dataset = Planetoid(root='gs://data/Planetoid', name='Cora', transform=NormalizeFeatures())

Suggest a potential alternative/fix

Either of two popular libraries can be used, fsspec or epath could be used to replace the existing calls to built-in functions that assume local filesystem.

I am happy to make a PR for this if I know the preference of filesystem abstraction library and any other concerns.

Contributor guide

Research direction
Replace core Python I/O functions (open, exists, is file, makedirs) with fsspec or epath equivalents, ensuring all file operations support remote filesystems. Test with various backends like gcs, s3, and local.
Tech stack
python
Domain
backendapiinfrastructure
Issue type
Refactor
DifficultyEstimated implementation difficulty for a new contributor, from 1 for very small changes to 5 for expert-level work.
3
Estimated timeA rough time range for an experienced contributor to investigate, implement, test, and prepare a pull request.
Half day
Activity statusHow available the issue appears right now: fresh, active, stale, blocked, or waiting on maintainer input.
Active
ClarityHow clearly the issue explains the expected change, acceptance criteria, and next step.
Mostly clear
Prerequisites
PythonPyTorchfsspec or epath
Newbie friendlinessA 1-100 score estimating how approachable this issue is for first-time contributors.
70