Datasets:

arxiv_dataset

like 67

At some point, this dataset's schema must have fallen out of sync with that of its source on Kaggle. Namely, the original dataset has versions and authors_parsed fields that are inaccessible here.

This PR adds those fields for completeness, and simply because they are useful. For example, arXiv papers in the GCS bucket are named by their version (e.g., 0704.0002v2.pdf), so the only way to download the latest one is to know that version. Currently, this cannot be done without listing all existing versions in the bucket. This PR would speed up and ease that process by providing the latest version as the last field in versions.

There are no storage drawbacks, since the entire Kaggle dataset has to be downloaded anyway, but the loaded dataset size in memory will increase.