iframe-proxy

AnneBeyer · 2026-01-06T14:55:00Z

Reference Issues/PRs

Fixes #32872 ~~and the predict part of #32866~~

What does this implement/fix? Explain your changes.

Previously, the case distinction for binary vs. multiclass was made based on the response dimension (see #32918 for reference), which led to multiclass predict ignoring the multiclass_colors parameter.
The colors are now selected based on the n_classes display class variable, which is extracted in from_estimator.

Note that a distinction based on the response shape still has to be made for plotting. ~~in two cases (contour and pcolormesh). I'm not sure if there is a more elegant solution, but I didn't find a way to handle pcolormesh in the same way for predict as in the other cases.~~

I adapted the docstrings and the exmple and added the response functions to the tests (and adapted them accordingly).

EDIT: The output of the code in issue #32872 now looks like this (Note that the levels problem from #32866 is still there for predict, but this PR is only concerned with making it use the same color map as the other methods, which it does now):

Since every class is plotted separately now (as it was already for response method predict_proba and decision_function), this also fixes the levels issue noted in #32866. However, there are white lines between the classes for predict with countourf. If this is a problem (and using pcolormesh is not an option for any reason), the current approach will need to be adapted further.

AI usage disclosure

I used AI assistance for:

Code generation (e.g., when writing an implementation or fixing a bug)
Test/benchmark generation
Documentation (including examples)
Research and understanding

Any other comments?

I also added a test for the default value for multiclass_colors, which was missing.

ogrisel · 2026-01-07T16:57:14Z

ogrisel

I didn't check the details but from a cursory glance this looks great. Please feel free to ping me once CI is green.

I think it would be great to also find a way to include a non regression test for #32866, that is, that different regions are assigned distinct colors when n_classes is large.

EDIT: maybe inspecting the levels would be a way to achieve this, for instance by including/adapting the test of #32866.

AnneBeyer · 2026-01-08T11:58:09Z

maybe inspecting the levels would be a way to achieve this, for instance by including/adapting the test of #32866.

For plot_method=contourf, the plotting is now handled differently, i.e., by adding a separate surface for each class, so the levels are not helping here. Instead, I added two separate checks that there are at least as many distinct colors as classes and that the number of surfaces matches the number of classes.

However, adding the level check part from PR #32867 when plot_method=contour revealed that the bug from #32866 is actually still a problem in this case and will require the adaption of levels proposed there. @leweex95, would you be interested in adapting that once this PR is merged?

AnneBeyer · 2026-01-08T11:59:29Z

-        "Blues",
-        ["red", "green", "blue"],
+        (None, 3),
+        (None, 15),


This test will fail for ("tab10", 15), which I will address in a follow-up PR.

AnneBeyer · 2026-01-08T12:01:32Z

+        ("plasma", 15),
+        ("Blues", 3),
+        ("Blues", 15),
+        (["red", "green", "blue"], 3),


I had to combine colors and number of classes here because the list version has to match the exact number of classes.

AnneBeyer · 2026-01-08T12:03:41Z

    else:
        assert_allclose(disp.surface_.colors, colors)

+    # non-regression test for issue #32866 with `contour` (currently still fails)


This can be fixed by integrating the approach proposed in #32867 once this is merged.

ogrisel

Here is an early pass of feedback with a focus on the meaning of this display and the "n_classes" params when we use this to display the output of the predict method for a regressor.

Co-authored-by: Olivier Grisel <olivier.grisel@ensta.org>

…ithub.com/AnneBeyer/scikit-learn into multiclass_colors_DecisionBoundaryDisplay

AnneBeyer · 2026-01-20T10:33:45Z

Thank you for all the comments @lucyleeow!

In order to keep this PR on point, I'll only address those related to making predict use multiclass_colors (and some minor fixes you suggested) here.

For the remaining points, which are equally valid, I'll try to summarize them here so we don't forget anything. Those should be addressed in follow-up PRs, though.

Add level handling, as already proposed here FIX: fixed DecisionBoundaryDisplay to display distinct colours for all classess correctly #32867 (once this one is merged)
Refactor the color handling (I opened a separate issue to keep track of that here: Fix and refactor color handling in DecisionBoundaryDisplay #33115)
Decide on how to handle contour (see RFC DecisionBoundaryDisplay levels overlap with 'contour' #33108)
Add support for clusterers (@ThexXTURBOXx, would you please open a separate issue for that?)
Add example for DecisionBoundary with regressor

Please add if I missed anything!

Co-authored-by: Lucy Liu <jliu176@gmail.com>

AnneBeyer · 2026-01-21T14:16:23Z

I think I addressed all comments @lucyleeow.
@ogrisel there have been some conceptual changes, could you please check if you still approve?

There are also a few remaining questions where feedback would be welcome.

…ithub.com/AnneBeyer/scikit-learn into multiclass_colors_DecisionBoundaryDisplay

ogrisel

I had another look at this PR and this LGTM as it is. It's already a big improvement. +1 for merge and tackling the remaining points discussed in linked issues in follow-up PRs.

ogrisel · 2026-01-30T09:44:25Z

@lucyleeow ok for merge?

lucyleeow

Thanks for all your work on this, it was quite complicated in the end.

Some nits only and a question but LGTM and fine to merge.

AnneBeyer · 2026-02-02T09:05:40Z

Thank you for all the feedback @lucyleeow and @ogrisel! I think now it is good to go.

lucyleeow · 2026-02-02T09:35:12Z

Thanks for your work and your patience @AnneBeyer ! Merging!

ogrisel · 2026-02-02T16:00:59Z

AnneBeyer added 4 commits December 17, 2025 16:24

pass n_classes to plot() via class instance

466d244

adapt code to use multiclass_colors for multiclass predict

68fcf6e

adapt docstrings and warning message

819eae2

adapt plotting function calls

46fbcaa

github-actions Bot added the module:inspection label Jan 6, 2026

add changelog

b0897df

AnneBeyer marked this pull request as draft January 6, 2026 15:34

AnneBeyer added 3 commits January 7, 2026 12:06

update changelog

48a3962

adapt failing tests

f3fdfca

fix docstring error

cdf8d4d

ogrisel reviewed Jan 7, 2026

View reviewed changes

Comment thread sklearn/inspection/_plot/tests/test_boundary_decision_display.py Outdated

StefanieSenger added this to Labs Jan 8, 2026

github-project-automation Bot moved this to Todo in Labs Jan 8, 2026

StefanieSenger moved this from Todo to In progress in Labs Jan 8, 2026

add non-regression tests for issue scikit-learn#32866

f58cad3

AnneBeyer commented Jan 8, 2026

View reviewed changes

AnneBeyer added 2 commits January 8, 2026 13:54

fix docstring errors

8080d8e

fix another docstring error

9e65d8c

AnneBeyer mentioned this pull request Jan 8, 2026

FIX: fixed DecisionBoundaryDisplay to display distinct colours for all classess correctly #32867

Closed

ogrisel mentioned this pull request Jan 8, 2026

ENH: Add xlim/ylim parameters to DecisionBoundaryDisplay.from_estimator #31693

Closed

ogrisel reviewed Jan 8, 2026

View reviewed changes

Comment thread sklearn/inspection/_plot/decision_boundary.py Outdated

Comment thread sklearn/inspection/_plot/decision_boundary.py Outdated

AnneBeyer and others added 4 commits January 9, 2026 11:49

fix indentation

c49c34f

Co-authored-by: Olivier Grisel <olivier.grisel@ensta.org>

adapt n_classes

5ae0ef0

Merge branch 'multiclass_colors_DecisionBoundaryDisplay' of https://g…

a4ed32e

…ithub.com/AnneBeyer/scikit-learn into multiclass_colors_DecisionBoundaryDisplay

add doc for regressor

86b8cbf

AnneBeyer marked this pull request as ready for review January 9, 2026 16:43

AnneBeyer commented Jan 20, 2026

View reviewed changes

Comment thread sklearn/inspection/_plot/decision_boundary.py Outdated

AnneBeyer mentioned this pull request Jan 20, 2026

Fix and refactor color handling in DecisionBoundaryDisplay #33115

Open

5 tasks

ThexXTURBOXx mentioned this pull request Jan 21, 2026

Support clusterers in DecisionBoundaryDisplay #33119

Open

AnneBeyer and others added 4 commits January 21, 2026 12:04

simplify code and adapt test and changelog

f34b1b0

adapt docstrings and add glossary links

09a7e1d

apply suggestion

f74178d

Co-authored-by: Lucy Liu <jliu176@gmail.com>

apply suggestion

73e12a0

AnneBeyer added 5 commits January 22, 2026 10:18

Merge branch 'multiclass_colors_DecisionBoundaryDisplay' of https://g…

b21b488

…ithub.com/AnneBeyer/scikit-learn into multiclass_colors_DecisionBoundaryDisplay

adapt docstrings and add versionchanged

3d8f051

fix copy-paste error

027ed0a

fix scatterplot color in example

72519a1

fix typo

f2b1f1d

StefanieSenger moved this from In progress to In progress - High Priority in Labs Jan 26, 2026

ogrisel approved these changes Jan 30, 2026

View reviewed changes

lucyleeow approved these changes Jan 31, 2026

View reviewed changes

Comment thread sklearn/inspection/_plot/tests/test_boundary_decision_display.py Outdated

Comment thread sklearn/inspection/_plot/decision_boundary.py

Comment thread sklearn/inspection/_plot/decision_boundary.py Outdated

update test and wording

1a2735e

lucyleeow merged commit 41f1f34 into scikit-learn:main Feb 2, 2026
40 checks passed

github-project-automation Bot moved this from In progress - High Priority to Done in Labs Feb 2, 2026

lesteve mentioned this pull request Feb 2, 2026

BUG undefined n_classes for custom estimators in DecisionBoundaryDisplay #33194

Closed

AnneBeyer deleted the multiclass_colors_DecisionBoundaryDisplay branch February 4, 2026 12:27

AnneBeyer mentioned this pull request Feb 17, 2026

FIX: Set levels explicitly to plot all classes in distinct colors in DecisionBoundaryDisplay #33300

Merged

4 tasks

StefanieSenger added this to Visualization and displays Feb 27, 2026

github-project-automation Bot moved this to Done in Visualization and displays Feb 27, 2026

jshn9515 mentioned this pull request Jul 3, 2026

Separate control over response surfaces and class boundaries in DecisionBoundaryDisplay #34431

Open

Sunbelt Computer Software

PL/B Language Development and Support

Uh oh!

Uh oh!

Conversation

AnneBeyer commented Jan 6, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Reference Issues/PRs

What does this implement/fix? Explain your changes.

AI usage disclosure

Any other comments?

Uh oh!

ogrisel commented Jan 7, 2026

Uh oh!

ogrisel left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

AnneBeyer commented Jan 8, 2026

Uh oh!

AnneBeyer Jan 8, 2026

Choose a reason for hiding this comment

Uh oh!

AnneBeyer Jan 8, 2026

Choose a reason for hiding this comment

Uh oh!

AnneBeyer Jan 8, 2026

Choose a reason for hiding this comment

Uh oh!

ogrisel left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

AnneBeyer commented Jan 20, 2026

Uh oh!

AnneBeyer commented Jan 21, 2026

Uh oh!

ogrisel left a comment

Choose a reason for hiding this comment

Uh oh!

ogrisel commented Jan 30, 2026

Uh oh!

lucyleeow left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

AnneBeyer commented Feb 2, 2026

Uh oh!

lucyleeow commented Feb 2, 2026

Uh oh!

Uh oh!

ogrisel commented Feb 2, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

AnneBeyer commented Jan 6, 2026 •

edited

Loading

ogrisel left a comment •

edited

Loading