[BUG] Error during fetch_annotation

[tincala91]

Issue summary

Cannot download any annotation due to path error: str object has no attribute mkdir, triggered in nilearn's fetch_single_file.

Apparently in version 0.11.0 of nilearn, they require a Path object for data_dir, while in fetch_annotation.py, l288 the data dir is casted to string.

Removing the str(data_dir) fixes issue, but it will be incompatible with older nilearn.

Detailed issue description

No response

Steps to reproduce issue

pip install -U nilearn=0.11.0

from pathlib import Path
p = Path(r'C:/Users/asala/neuromaps-data')

from neuromaps.datasets import fetch_annotation
from neuromaps.datasets import available_annotations
for annotation in available_annotations(tags='PET'):
    print(annotation)
    fetch_annotation(source=annotation[0], desc=annotation[1],
                     space=annotation[2], den=annotation[3],
                     data_dir=p)

Software version

No response

Code of Conduct

• I agree to follow the neuromaps Code of Conduct

[liuzhenqi77]

Hi, thanks so much for reporting and pinpointing the bug! I've opened a PR for this.

Removing the str() works for me for both 0.11.0 and 0.10.X versions. May I ask which older nilearn version would be incompatible on your end? Thanks!

[tincala91]

I checked with nilearn 0.10.4, and it accepts the string object instead of path, and use the usual os.path functions.

os.path accepts Path objects since Python3.6, so I feel it is actually safe to remove the cast to string.