Christopher Akiki
Christopher Akiki
@albertvillanova Adding the validation split causes these two `assert_called_once` assertions to fail with `AssertionError: Expected 'ArrowWriter' to have been called once. Called 2 times`: https://github.com/huggingface/datasets/blob/main/tests/test_builder.py#L548-L562 It might be better to...
Streaming with `split=all` seems to be working, will fix the failing test next
Not sure if marking the PR as "ready for review" actually notified you, so tagging @albertvillanova just in case :smiley_cat:
cc @lhoestq
@albertvillanova Nice! Let me know if it's something I can fix my self; would love to contribtue!
Hi @albertvillanova ! Sorry it took so long; I wanted to spend this weekend working on it.
Happy to work on this if @PierreColombo does not have time!
This might be a tensorflow issue, but should have been fixed in https://github.com/huggingface/evaluate/commit/9b6cea3ef7f84dd619d65a6fd5a7da07f2386fae afaik. What version of evaluate are you using?
@lewtun I'd like to help with this!
@ola13 Yes :smile_cat: ! The ones I'm adding are ranked relevance metrics like nDCG@k and MAP@k, so not sure they're super useful for retrieval on ROOTS with no relevance judgments....