Ridge Regressor Precision Errors

[Oddant1]

Bug Description
The Ridge Regressor tests are off by relatively small amounts in some environments causing test failure. Busywork is unaffected, and @nbokulich was unable to reproduce the failures. @thermokarst and I got the test failures on qiime2-2020.2, qiime2-2020.6, and latest dev (as of this issue opening) environments.

Steps to reproduce the behavior

1. Run the q2-sample-classifier tests (run using py.test --pyargs q2_sample_classifier not make test if attempting to test a specific conda environment and not currently checked out source)
2. Possibly observe something resembling the below screenshot

Expected behavior
The tests should pass in all environments

Screenshots

Computation Environment
Myself

• OS: Linux Mint 19.3
• QIIME 2: 2020.2, 2020.6, latest dev

@thermokarst

• OS: Darwin
• QIIME 2: 2020.2, 2020.6

Comments

1. @nbokulich said this is a pretty minor issue due to the relatively small discrepancies between result and expected result as well as the fact the Ridge Regressor is unused.
2. Possibly caused by recent sklearn changes to Ridge (see reff).
3. Query why this issue only occurs in some environments.

References

1. https://scikit-learn.org/stable/whats_new/v0.23.html