Datasets:
Three column names contain the equal sign
Three of the column names contain the = sign:
Language=License=Meta_Link=
Please note that column names should only contain alphanumeric characters or hyphen or underscore.
Note that the presence of the equal sign may break some functionalities of the dataset viewer, for example filtering a column by one of its values:
- It raises an error when filtering
Language=column withenvalue: https://huggingface.co/datasets/m-a-p/Matrix/viewer/default/train?f[Language=][value]=%27en%27 - Whereas it works when filtering
ocaidcolumn withnullvalue: https://huggingface.co/datasets/m-a-p/Matrix/viewer/default/train?f[ocaid]=null
Is there a way to fix that on the backend of the viewer ? It might not be easy to reupload this TBs of data just for that (though it could be a good opportunity to compress the data)
Hi @lhoestq ,
Certainly we could make the backend more robust against this kind of potential issues. I am proposing a PR in this sense.
Nevertheless, I think this specific case is a data quality issue caused by a bug on their side while processing the data. It is a good practice to make some quality checks on the datasets before uploading them to the Hub. By regenerating the data, they could fix the inconsistency in the column names (ones with trailing equal sign character and other without it), as well as improve downloading performance (and disk store usage) by compressing their data.
Fixed! Thanks @albertvillanova for working on this.