FIX Do not use deprecated API in fetch_20newsgroups_vectorized #21216
Conversation
| "col1": pd.arrays.SparseArray([0, 1, 0], dtype=ntype1, fill_value=0), | ||
| "col2": pd.arrays.SparseArray([1, 0, 1], dtype=ntype2, fill_value=0), |
There was a problem hiding this comment.
These are causing issues on pandas because the fill value is different depending on the dtype.
And this is raising a ValueError on the dev version of pandas
There was a problem hiding this comment.
Does your change then fix the issue? Can't tell from your explanation
There was a problem hiding this comment.
I would guess that this fixes the issue but to double-check I pushed an empty commit with [scipy-dev] to trigger the scipy-dev nightly build in this PR.
There was a problem hiding this comment.
Yea it fixes the original issue.
There were two issues. With the fix to fetch_20newsgroups_vectorized, it alloed the tests to run, which revealed the pandas issue.
|
Also @thomasjpfan I am quite curious how you made the connection between the original cryptic pytest error and |
Turned off pytest-xdist by setting |
…t-learn#21216) * FIX Do not use deprecated API in fetch_20newsgroups_vectorized * BLD [scipy-dev] * TST Be explicit about fill value [scipy-dev] * TST Fixes tests for fill value * [scipy-dev] trigger nightly build Co-authored-by: Loïc Estève <loic.esteve@ymail.com>
* FIX Do not use deprecated API in fetch_20newsgroups_vectorized * BLD [scipy-dev] * TST Be explicit about fill value [scipy-dev] * TST Fixes tests for fill value * [scipy-dev] trigger nightly build Co-authored-by: Loïc Estève <loic.esteve@ymail.com>
…t-learn#21216) * FIX Do not use deprecated API in fetch_20newsgroups_vectorized * BLD [scipy-dev] * TST Be explicit about fill value [scipy-dev] * TST Fixes tests for fill value * [scipy-dev] trigger nightly build Co-authored-by: Loïc Estève <loic.esteve@ymail.com>
Reference Issues/PRs
Fixes #21212
What does this implement/fix? Explain your changes.
Errored during test collection time because of the deprecation warning.
Any other comments?
During test collection time and
SKLEARN_SKIP_NETWORK_TESTS='0', datasets are downloaded first to cache them. This is done because when multiple processes callfetch_*they can both write to the same location on the filesystem, which use to cause errors.