Can't load the dataset
#1
by
SaulLu
- opened
It seems that this dataset can only be used on JZ. What is the reason?
Traceback (most recent call last):
File "/home/lucile/repos/m4/m4/evaluation/scripts/create_sample_evaluation_datasets_simplified.py", line 15, in <module>
ds = load_dataset(name_ds, use_auth_token=True)
File "/home/lucile/anaconda3/envs/m4-user/lib/python3.8/site-packages/datasets/load.py", line 1718, in load_dataset
builder_instance = load_dataset_builder(
File "/home/lucile/anaconda3/envs/m4-user/lib/python3.8/site-packages/datasets/load.py", line 1498, in load_dataset_builder
builder_cls = import_main_class(dataset_module.module_path)
File "/home/lucile/anaconda3/envs/m4-user/lib/python3.8/site-packages/datasets/load.py", line 115, in import_main_class
module = importlib.import_module(module_path)
File "/home/lucile/anaconda3/envs/m4-user/lib/python3.8/importlib/__init__.py", line 127, in import_module
return _bootstrap._gcd_import(name[level:], package, level)
File "<frozen importlib._bootstrap>", line 1014, in _gcd_import
File "<frozen importlib._bootstrap>", line 991, in _find_and_load
File "<frozen importlib._bootstrap>", line 975, in _find_and_load_unlocked
File "<frozen importlib._bootstrap>", line 671, in _load_unlocked
File "<frozen importlib._bootstrap_external>", line 843, in exec_module
File "<frozen importlib._bootstrap>", line 219, in _call_with_frames_removed
File "/home/lucile/.cache/huggingface/modules/datasets_modules/datasets/HuggingFaceM4--SNLI-VE/fea4d17ddaf22ae6847bb7ae82937ba42f8ec9604e2048c4785fb0a6f28168a9/SNLI-VE.py", line 62, in <module>
JZ_FOLDER_PATH = f"{os.environ['cnw_ALL_CCFRSCRATCH']}/local_datasets/flickr30k-images.tar.gz"
File "/home/lucile/anaconda3/envs/m4-user/lib/python3.8/os.py", line 675, in __getitem__
raise KeyError(key) from None
KeyError: 'cnw_ALL_CCFRSCRATCH'
The issue is that the flickr dataset has to be manually downloaded. Initially I thought it'd be better to have the path directly in the script. However, since we want to use it outside of JZ, I switched to using a manual directory. Only issue is that it has to be specified every time we use the dataset.
manual_download_dir is the right way to go imo (especially when we switch this to public dataset).
can you add the manual_download_instructions property for clarity?
See https://huggingface.co/datasets/story_cloze/blob/main/story_cloze.py#L50 for example
yes for sure! I took it off when I switched to the jz_path, but I'll add it back