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

leweex95 · 2025-12-08T20:20:03Z

This PR aims to fix a bug where DecisionBoundaryDisplay.from_estimator only displayed up to 7 distinct colors for classifiers with more than 7 classes when using plot_method="contourf" or plot_method="contour"

Fixes #32866

…l classes correctly.

ThexXTURBOXx

This fixes the linked issue - no problems remain. Thank you very much!

StefanieSenger · 2025-12-09T08:23:54Z

Thanks for the PR, @leweex95!

@AnneBeyer, since you are concerned with Displays, would you be interested in triaging the corresponding issue and reviewing this PR?

ogrisel

I am assessing the impact of this PR visually using the following code:

Code

import numpy as np
import matplotlib.pyplot as plt
from sklearn.linear_model import LogisticRegression
from sklearn.inspection import DecisionBoundaryDisplay


data = np.array(
    [
        [-1, -1],
        [-2, -1],
        [1, 1],
        [2, 1],
        [2, 2],
        [3, 2],
        [3, 3],
        [4, 3],
        [4, 4],
        [5, 4],
        [5, 5],
    ]
)
# target = np.asarray([str(i) for i in range(11)])
target = np.arange(11)
clf = LogisticRegression().fit(data, target)

cmap = "gist_rainbow"
_, axes = plt.subplots(nrows=3, ncols=3, figsize=(12, 12), constrained_layout=True)
for plot_method_idx, plot_method in enumerate(["contourf", "contour", "pcolormesh"]):
    for response_method_idx, response_method in enumerate(
        ["predict_proba", "decision_function", "predict"]
    ):
        ax = axes[plot_method_idx, response_method_idx]
        display = DecisionBoundaryDisplay.from_estimator(
            clf,
            data,
            multiclass_colors=cmap,
            response_method=response_method,
            plot_method=plot_method,
            ax=ax,
            alpha=0.5,
        )
        ax.scatter(
            data[:, 0],
            data[:, 1],
            c=target.astype(int),
            edgecolors="black",
            cmap=cmap,
        )
        if plot_method != "pcolormesh":
            if isinstance(display.surface_, list):
                levels = len(display.surface_[0].levels)
            else:
                levels = len(display.surface_.levels)
            ax.set_title(f"plot_method={plot_method}\nresponse_method={response_method}\nlevels={levels}")
        else:
            ax.set_title(f"plot_method={plot_method}\nresponse_method={response_method}")

Results

`main`

this branch

Analysis

So I confirm that this fixes two bugs:

one for contour plot method in conjunction with the predict response method;
one for contourf plot method in conjunction with the predict_proba and decision_function response methods.

I think the changelog should be adjusted to better reflect this.

See also some further suggestions to improve the tests.

ogrisel · 2025-12-09T09:39:50Z

sklearn/inspection/_plot/tests/test_boundary_decision_display.py

+            [5, 5],
+        ]
+    )
+    y = np.arange(11)


Could you please parametrize this test to make sure that it still works as expected when the dtype of y is object with str instances as class labels?

I think the code in this PR should work since from_estimator always uses a LabelEncoder but better cover this case in this test.

ogrisel · 2025-12-09T10:04:43Z

sklearn/inspection/_plot/tests/test_boundary_decision_display.py

+    clf = LogisticRegression().fit(pts, y)
+
+    disp = DecisionBoundaryDisplay.from_estimator(
+        clf, pts, response_method="predict", plot_method=plot_method


Please extend this test for predict_proba and decision_function as well, at least for the "contour".

ogrisel · 2025-12-09T13:29:42Z

Maybe @lucyleeow would also be interested in taking a look at this PR given her past involvement with this class.

AnneBeyer

Thanks for the PR, @leweex95!

In general, it looks good to me, I just added a few comments below.

AnneBeyer · 2025-12-09T13:09:04Z

sklearn/inspection/_plot/decision_boundary.py

+                    )
+                else:
+                    levels = unique_levels
+                kwargs.setdefault("levels", levels)


I'm not sure if this is too defensive (I'm not too familiar with the use of levels), but since the user can also pass kwargs (like levels) directly to from_estimator, maybe this should only be computed and added if levels is not given?

Good point. And the test should be updated accordingly.

AnneBeyer · 2025-12-09T13:10:12Z

sklearn/inspection/_plot/decision_boundary.py

                # Plot only argmax map for contour
                class_map = self.response.argmax(axis=2)
+                # Ensure levels match the number of classes for distinct colors
+                kwargs.setdefault("levels", np.unique(class_map))


Same as above, only set if not given?

AnneBeyer · 2025-12-09T13:15:54Z

sklearn/inspection/_plot/tests/test_boundary_decision_display.py

+    )
+
+    # Check that the surface has levels for all classes
+    if plot_method in ("contourf", "contour"):


plot_method can only be either "contourf" or "contour" (as defined by parametrize), so I don't think this check is needed.

I can also be set to "pcolormesh" (see the code snippet I posted above) but maybe this is not consistently documented?

In general, yes, but this test is (at least currently) only testing "contourf" or "contour". So if the test is extended, it should stay, otherwise, it is not necessary here.

Ah indeed, sorry for the noisy comment. Since this test is about testing the actual number of levels which is undefined for "pcolormesh", let's remove this condition.

leweex95 · 2025-12-13T12:10:46Z

Thanks for your comments, @AnneBeyer and @ogrisel , pushed the fixes!

lucyleeow · 2025-12-15T04:29:32Z

Thanks for your PR @leweex95, I think it would be best if #32872 was addressed first (see #32872 (comment) for details) - that fix would probably remove the binary (and predict)/multiclass logic we have currently for color handling, and also it would probably pass n_classes information, so we can avoid any expensive np.unique calls.

@AnneBeyer is going to work on #32872. You're welcome to pick this up after #32872 is fixed (there may be some significant merge conflicts though, so you may find it easier to start a new PR).

Edit: Note for the new PR I think we should add a TODO to delete this extra logic once our min matplotlib version is >=3.11 since it has been fixed in matplotlib/matplotlib#27576

FIX: fixed DecisionBoundaryDisplay to display distinct colours for al…

0f48349

…l classes correctly.

github-actions bot added the module:inspection label Dec 8, 2025

leweex95 mentioned this pull request Dec 8, 2025

DecisionBoundaryDisplay.from_estimator only displays up to 7 distinct colours #32866

Open

github-actions bot added the CI:Linter failure The linter CI is failing on this PR label Dec 8, 2025

fixed ruff-detected formatting issues

f1ad158

github-actions bot removed the CI:Linter failure The linter CI is failing on this PR label Dec 8, 2025

leweex95 and others added 2 commits December 8, 2025 21:42

updated whatsnew .rst file

86c4c16

Merge branch 'main' into fix-decision-boundary-colors

d0d3945

ThexXTURBOXx approved these changes Dec 9, 2025

View reviewed changes

ogrisel reviewed Dec 9, 2025

View reviewed changes

AnneBeyer reviewed Dec 9, 2025

View reviewed changes

AnneBeyer mentioned this pull request Dec 9, 2025

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

Open

ogrisel mentioned this pull request Dec 9, 2025

DecisionBoundaryDisplay with response_method="predict" has inconsistent handling for the colormap in the multiclass case #32872

Open

leweex95 added 2 commits December 13, 2025 13:09

updating changelog

0ef5152

addressing review comments

e79d404

Merge branch 'main' into fix-decision-boundary-colors

41e743d

Uh oh!

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

Are you sure you want to change the base?

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

Conversation

leweex95 commented Dec 8, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

ThexXTURBOXx left a comment

Choose a reason for hiding this comment

Uh oh!

StefanieSenger commented Dec 9, 2025

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

Code

Results

main

this branch

Analysis

Uh oh!

ogrisel Dec 9, 2025

Choose a reason for hiding this comment

Uh oh!

ogrisel Dec 9, 2025

Choose a reason for hiding this comment

Uh oh!

ogrisel commented Dec 9, 2025

Uh oh!

AnneBeyer left a comment

Choose a reason for hiding this comment

Uh oh!

AnneBeyer Dec 9, 2025

Choose a reason for hiding this comment

Uh oh!

ogrisel Dec 9, 2025

Choose a reason for hiding this comment

Uh oh!

AnneBeyer Dec 9, 2025

Choose a reason for hiding this comment

Uh oh!

AnneBeyer Dec 9, 2025

Choose a reason for hiding this comment

Uh oh!

ogrisel Dec 9, 2025 • 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!

AnneBeyer Dec 9, 2025

Choose a reason for hiding this comment

Uh oh!

ogrisel Dec 9, 2025 • 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!

leweex95 commented Dec 13, 2025

Uh oh!

lucyleeow commented Dec 15, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

6 participants

leweex95 commented Dec 8, 2025 •

edited

Loading

ogrisel left a comment •

edited

Loading

`main`

ogrisel Dec 9, 2025 •

edited

Loading

ogrisel Dec 9, 2025 •

edited

Loading

lucyleeow commented Dec 15, 2025 •

edited

Loading