Bug#1019511: scipy breaks scikit-learn autopkgtest on ppc64el: precision delta's

[Paul Gevers]

Source: scipy, scikit-learn
Control: found -1 scipy/1.8.1-14
Control: found -1 scikit-learn/1.1.2+dfsg-5
Severity: serious
Tags: sid bookworm
User: debian-ci at lists.debian.org
Usertags: breaks needs-update

Dear maintainer(s),

With a recent upload of scipy the autopkgtest of scikit-learn fails in 
testing when that autopkgtest is run with the binary packages of scipy 
from unstable. It passes when run with only packages from testing. In 
tabular form:

                        pass            fail
scipy                  from testing    1.8.1-14
scikit-learn           from testing    1.1.2+dfsg-5
all others             from testing    from testing

I copied some of the output at the bottom of this report.

Currently this regression is blocking the migration of scipy to testing 
[1]. Due to the nature of this issue, I filed this bug report against 
both packages. Can you please investigate the situation and reassign the 
bug to the right package?

More information about this bug and the reason for filing it can be found on
https://wiki.debian.org/ContinuousIntegration/RegressionEmailInformation

Paul

[1] https://qa.debian.org/excuses.php?package=scipy

https://ci.debian.net/data/autopkgtest/testing/ppc64el/s/scikit-learn/25863055/log.gz

=================================== FAILURES 
===================================
______________________ test_mlp_regressor_dtypes_casting 
_______________________

     def test_mlp_regressor_dtypes_casting():
         mlp_64 = MLPRegressor(
             alpha=1e-5, hidden_layer_sizes=(5, 3), random_state=1, 
max_iter=50
         )
         mlp_64.fit(X_digits[:300], y_digits[:300])
         pred_64 = mlp_64.predict(X_digits[300:])
             mlp_32 = MLPRegressor(
             alpha=1e-5, hidden_layer_sizes=(5, 3), random_state=1, 
max_iter=50
         )
         mlp_32.fit(X_digits[:300].astype(np.float32), y_digits[:300])
         pred_32 = mlp_32.predict(X_digits[300:].astype(np.float32))
     >       assert_allclose(pred_64, pred_32, rtol=1e-04)
E       AssertionError: E       Not equal to tolerance rtol=0.0001, atol=0
E       E       Mismatched elements: 1 / 60 (1.67%)
E       Max absolute difference: 1.77346709e-06
E       Max relative difference: 0.00013333
E        x: array([-1.624248e-02,  2.327707e+00,  6.674963e-01, 
4.904700e-01,
E               6.739288e-01,  3.166697e+00,  4.548126e-01,  6.674963e-01,
E              -3.220949e-02, -6.899952e-01,  6.674963e-01, 
-6.329127e-01,...
E        y: array([-1.624250e-02,  2.327706e+00,  6.674960e-01, 
4.904711e-01,
E               6.739284e-01,  3.166698e+00,  4.548138e-01,  6.674960e-01,
E              -3.220773e-02, -6.899955e-01,  6.674960e-01, 
-6.329128e-01,...

/usr/lib/python3/dist-packages/sklearn/neural_network/tests/test_mlp.py:872: 
AssertionError
-------------- next part --------------
A non-text attachment was scrubbed...
Name: OpenPGP_signature
Type: application/pgp-signature
Size: 495 bytes
Desc: OpenPGP digital signature
URL: <http://alioth-lists.debian.net/pipermail/debian-science-maintainers/attachments/20220910/548cd156/attachment.sig>