Dataset incomplete for fields like anchors->wikipedia_id
The ['anchors']['wikipedia_id'] only exists for a small part of the dataset.
For example, the result of full_dataset[0]['anchors']['wikipedia_id'] is ['378194', '3675310', '32693', ...].
The result of full_dataset[-1]['anchors']['wikipedia_id'] is instead ['', '', '', ...]
I checked the dataset item by item, and found out that the wikipedia_id in anchors is invalid for 5508446 out of 5903530 items (93.3% are invalid).
I understand that this column is not important for most of the users. But is there any way that I can get the full dataset, or is the code for generating http://dl.fbaipublicfiles.com/KILT/kilt_knowledgesource.json public so that I can convert a dump to dataset by my own?
Thanks in advance :)
Hi @drt , thanks for reporting.
Please note that at Hugging Face we only host a loading script which downloads the original data file created by the authors from: http://dl.fbaipublicfiles.com/KILT/kilt_knowledgesource.json
If there are issues with the data itself, you should better contact the authors/owners of the data: Facebook Research.
Maybe you could open an issue in their GitHub repository for the KILT dataset: https://github.com/facebookresearch/KILT
- I have seen you commented this issue: https://github.com/facebookresearch/KILT/issues/62
Thanks a lot, @albertvillanova :)
Yes, the problem is in the dataset itself, I posted a walk around in the GitHub issue page.