List of improvements expected for tslearn version 0.6.2

[YannCabanes]

For the version 0.6.2 of tslearn, the following improvements are expected:

• Fix the test error with MacOS for Python 3.9 (Segmentation fault):
  • test_symmetric_cdist --> The failing rate of this test is about 3/4
  • test_gamma_soft_dtw --> The failing rate of this test is about 1/5
    When these tests are failing, the error messages can be different from one fail to the other: the Segmentation fault can occur in different lines of the code. These two tests are currently skipped with MacOS for Python 3.9.
• In the tests, do we have to specify the backend if our data is prepared using be?
  We should test both cases: with backend inferred from data and with backend explicitly specified, even with data from another format.
  --> This problem has been dealt with in PR Cast the input data to backend option and test functions in metrics for different input data types and different backend options. #479.
• The documentation about the backends should be improved to explain how they work.
  For example what does it mean for DTW to support the PyTorch backend, do we differentiate at fixed path?
  --> This problem has been dealt with in PR Improve backend docstrings #482.
• The documentation of the class SoftDTWLossPyTorch should be improved.
  --> The documentation of the class SoftDTWLossPyTorch has been improved in the PR Improve the docstring of the class SoftDTWLossPyTorch #467.
• For the class Backend in backend/backend.py, should we overload __getattr__ instead of listing all attributes?
  --> This requirement has been dealt with in PR Simplify backend #471.

[codecov-commenter]

Codecov Report

Patch and project coverage have no change.

Comparison is base (f718dd5) 93.06% compared to head (d375389) 93.06%.

❗ Your organization is not using the GitHub App Integration. As a result you may experience degraded service beginning May 15th. Please install the Github App Integration for your organization. Read more.

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #460   +/-   ##
=======================================
  Coverage   93.06%   93.06%           
=======================================
  Files          67       67           
  Lines        5625     5625           
=======================================
  Hits         5235     5235           
  Misses        390      390           

☔ View full report in Codecov by Sentry.
📢 Do you have feedback about the report comment? Let us know in this issue.

[rtavenar]

If I may add something: at the moment, the tests install scikit-learn v1.0, which does not provide wheels for Python 3.10. As a result, for all tests running on Python 3.10, a large part of the test time is dedicated to compiling scikit-learn. It would be a good idea, if feasible, to switch to scikit-learn 1.3 for the tests.

[YannCabanes]

The documentation of the class SoftDTWLossPyTorch has been improved in the PR #467.

[rtavenar]

If I may add something: at the moment, the tests install scikit-learn v1.0, which does not provide wheels for Python 3.10. As a result, for all tests running on Python 3.10, a large part of the test time is dedicated to compiling scikit-learn. It would be a good idea, if feasible, to switch to scikit-learn 1.3 for the tests.

We are not there (=1.3) yet, however we upgraded tests to support scikit-learn 1.2, which provides wheels for Python 3.8 to 3.11 it seems