MNT Deprecate SAMME.R algorithm from AdaBoostClassifier

[StefanieSenger]

Reference Issues/PRs

Fixes #26784

What does this implement/fix? Explain your changes.

This PR deprecates the SAMME.R algorithm as an algorithm in the AdaBoostClassifier as discussed in issue #26784.

[github-actions]

✔️ Linting Passed

All linting checks passed. Your pull request is in excellent shape! ☀️

_{Generated for commit: 0e5656e. Link to the linter CI: here}

[adrinjalali]

otherwise LGTM.

[adrinjalali]

cc @glemaitre

[adrinjalali]

Could you also make sure this parameter is set in all examples correctly so that examples and our documentation won't raise a warning?

[glemaitre]

The changes look good. We need to make the changes suggested by @adrinjalali.

I think that we should remove the following example: https://scikit-learn.org/dev/auto_examples/ensemble/plot_adaboost_hastie_10_2.html#sphx-glr-auto-examples-ensemble-plot-adaboost-hastie-10-2-py. We need to change the user guide figure and probably redirect this example to the one that you edited earlier.

The redirection is done in the doc/conf.py file:

scikit-learn/doc/conf.py

Lines 283 to 302 in 75f3e47

	# redirects dictionary maps from old links to new links
	redirects = {
	"documentation": "index",
	"auto_examples/feature_selection/plot_permutation_test_for_classification": (
	"auto_examples/model_selection/plot_permutation_tests_for_classification"
	),
	"modules/model_persistence": "model_persistence",
	"auto_examples/linear_model/plot_bayesian_ridge": (
	"auto_examples/linear_model/plot_ard"
	),
	"auto_examples/model_selection/grid_search_text_feature_extraction.py": (
	"auto_examples/model_selection/plot_grid_search_text_feature_extraction.py"
	),
	"auto_examples/miscellaneous/plot_changed_only_pprint_parameter": (
	"auto_examples/miscellaneous/plot_estimator_representation"
	),
	"auto_examples/decomposition/plot_beta_divergence": (
	"auto_examples/applications/plot_topics_extraction_with_nmf_lda"
	),
	}

we need to map the previous example name to the new one.

[StefanieSenger]

Thank you, @glemaitre and @adrinjalali, for the review and the directions. I have tried to implement this; lets hope most of it correctly.

Some comments about some things I was unsure about:

I deleted the whole example plot-adaboost-hastie-10-2-py. Is that correct?

I changed where AdaBoostClassifier was instantiated with the default algorithms to now use “SAMME”, but was not sure about benchmarks/bench_20newsgroups.py, this one should stay as it is, right?

I modified the expected output lines of the example in the docstring of AdaBoostClassifier by hand so that it passes the doctest. Is this the correct way of dealing with this? I investigating how to run this automatic, but I didn’t manage to find out.

In the user guide (doc/modules/ensemble.rst), li. 1580-1581 needs to be changed in version 1.6. How would we document this, so it’s not forgotten (since .. /n TODO(1.6) will again show up on IDEs?)? Those two lines I mean:

For multi-class classification, :class:`AdaBoostClassifier` implements
AdaBoost-SAMME and AdaBoost-SAMME.R [ZZRH2009]_.

Looking forward to your further reviews!

[adrinjalali]

I deleted the whole example plot-adaboost-hastie-10-2-py. Is that correct?

Sounds good.

I changed where AdaBoostClassifier was instantiated with the default algorithms to now use “SAMME”, but was not sure about benchmarks/bench_20newsgroups.py, this one should stay as it is, right?

I think we should fix the benchmark as well.

I modified the expected output lines of the example in the docstring of AdaBoostClassifier by hand so that it passes the doctest. Is this the correct way of dealing with this? I investigating how to run this automatic, but I didn’t manage to find out.

We basically copy the code to a python REPL (not ipython),

In the user guide (doc/modules/ensemble.rst), li. 1580-1581 needs to be changed in version 1.6. How would we document this, so it’s not forgotten (since .. /n TODO(1.6) will again show up on IDEs?)? Those two lines I mean:

For multi-class classification, :class:`AdaBoostClassifier` implements
AdaBoost-SAMME and AdaBoost-SAMME.R [ZZRH2009]_.

We can already remove SAMME.R from the docs, our user guides shouldn't mention something which is already deprecated.

[StefanieSenger]

I believe the tests will only be changed, once SAMME.R is entirely abandoned in 1.6, right?

[glemaitre]

Some of the failures are still legitimate. You can reproduce the error locally by running pytest with the following option: -Werror::FutureWarning:

pytest -vsl sklearn -Werror::FutureWarning

Regarding the fixes:

• you need to either set algorithm="SAMME" when it makes sense
• or decorate a test to filter the warning using @pytest.mark.filterwarnings("ignore:The SAMME.R algorithm") and add a # TODO(remove in 1.6) to remember to remove and change the test in the future.

For the docstring test, we will need to modify the instantiation of AdaBoostClassifier to use "SAMME".

[glemaitre]

@StefanieSenger would you mind resolving the conflicts? I can make a final review after.

[StefanieSenger]

Sure, @glemaitre, I have just resolved the conflicts. :) I hadn't noticed them before.

[glemaitre]

LGTM. I merge main in this branch. Let see if the CI will pass.

[StefanieSenger]

Thanks, @glemaitre. The CI failure is not related, I believe?

[glemaitre]

Nop something link to GBDT and the new loss. We are tracking there: #27312

[glemaitre]

Merging then.

[glemaitre]

Thanks @StefanieSenger