huggingface / huggingface/datasets

load_from_disk with remote filesystem fails due to a wrong temporary local folder path

Open
#5,114 2 comments 0 reactions 0 assignees View on GitHub
bug
Dominant language
Python
Stars
22k
Forks
3.4k
Avg merge
5d 7h
Merged PRs (30d)
17

Description

## Describe the bug

The function load_from_disk fails when using a remote filesystem because of a wrong temporary path generation in the load_from_disk method of arrow_dataset.py:

```python
if is_remote_filesystem(fs):
src_dataset_path = extract_path_from_uri(dataset_path)
dataset_path = Dataset._build_local_temp_path(src_dataset_path)
fs.download(src_dataset_path, dataset_path.as_posix(), recursive=True)
```
If _dataset_path_ is `gs://speech/mydataset/train`, then _src_dataset_path_ will be `speech/mydataset/train` and _dataset_path_ will be something like `/var/folders/9s/gf0b/T/tmp6t/speech/mydataset/train`

Then, after downloading the **folder** _src_dataset_path_, you will get a path like `/var/folders/9s/gf0b/T/tmp6t/speech/mydataset/train/train/state.json` (notice we have train twice)

Instead of downloading the remote folder we should be downloading all the files in the folder for the path to be right:

```python
fs.download(os.path.join(src_dataset_path,*), dataset_path.as_posix(), recursive=True)
```

## Steps to reproduce the bug
```python
fs = gcsfs.GCSFileSystem(**storage_options)
dataset = load_from_disk("common_voice_processed") # loading local dataset previously saved locally, works fine
dataset.save_to_disk(output_dir, fs=fs) #works fine
dataset = load_from_disk(output_dir, fs=fs) # crashes
```

## Expected results
The dataset is loaded

## Actual results

FileNotFoundError: [Errno 2] No such file or directory: '/var/folders/9s/gf0b9jz15d517yrf7m3nvlxr0000gn/T/tmp6t5e221_/speech/datasets/tests/common_voice_processed/train/state.json'

## Environment info

- `datasets` version: datasets-2.6.1.dev0
- Platform: mac os monterey 12.5.1
- Python version: 3.8.13
- PyArrow version:pyarrow==9.0.0

Contributor guide

Open the contributing guide

Assessment

This issue has not been assessed yet.

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.