search

rapidsai / cuml

cuML - RAPIDS Machine Learning Library
https://docs.rapids.ai/api/cuml/stable/
Apache License 2.0
4.25k stars 534 forks source link

[DO NOT MERGE] Check that full CI tests are NumPy 2 ready #6029

Closed seberg closed 3 months ago

seberg commented 3 months ago

CuPy 13.3 is coming soon, and that should mean we can simply remove the numpy<2.0 pin. However, while we tested quite extensively locally, I have never made sure that the full CI should be passing just fine.

This tests CI, with CuPy 13.2.0, but applying the important patches (via monkeypatching). It also unpins pynndescent since it has (trivial) issues with NumPy 2.

Not meant to be merged, simply a double check, to see if there might be any need for fix-ups.

seberg commented 3 months ago

@jakirkham ping just FYI. I consider this a success. There is one nltk lookup error that looks like it is due to a connection error. So unrelated.

One note, I tried locally with CuPy 13.3.0 "beta" release. It contains a PR improving random.choice() which changes the stream. This affects a couple of tests:

FAILED test_doctest.py::test_docstring[NearestNeighbors] - AssertionError: 1 of 9 doctests failed for NearestNeighbors:
FAILED test_doctest.py::test_docstring[make_classification] - AssertionError: 1 of 3 doctests failed for make_classification:
FAILED test_doctest.py::test_docstring[MulticlassClassifier] - AssertionError: 1 of 7 doctests failed for MulticlassClassifier:
FAILED test_doctest.py::test_docstring[OneVsOneClassifier] - AssertionError: 1 of 7 doctests failed for OneVsOneClassifier:
FAILED test_doctest.py::test_docstring[OneVsRestClassifier] - AssertionError: 1 of 7 doctests failed for OneVsRestClassifier:

I assume it will be easy to adjust the tests and skip them when running on older CuPy versions.