ENH Add multi-threshold classification to FixedThresholdClassifier #31544
Conversation
|
Tests failing. Once CI is green feel free to ping me back. |
72c5623 to
05bc405
Compare
|
Hi @adrinjalali . "sklearn/model_selection/_classification_threshold.py:docstring of sklearn.model_selection._classification_threshold.FixedThresholdClassifier:105: WARNING: Block quote ends without a blank line; unexpected unindent. [docutils]" |
|
This should fix the issue: diff --git a/sklearn/model_selection/_classification_threshold.py b/sklearn/model_selection/_classification_threshold.py
index ab8ab266f4..37c1380cd8 100644
--- a/sklearn/model_selection/_classification_threshold.py
+++ b/sklearn/model_selection/_classification_threshold.py
@@ -293,7 +293,7 @@ class FixedThresholdClassifier(BaseThresholdClassifier):
>>> from sklearn.metrics import confusion_matrix
>>> from sklearn.model_selection import FixedThresholdClassifier, train_test_split
- # Binary classification with custom threshold
+ >>> # Binary classification with custom threshold
>>> X, y = make_classification(
... n_samples=1_000, weights=[0.9, 0.1], class_sep=0.8, random_state=42
... )
@@ -311,13 +311,13 @@ class FixedThresholdClassifier(BaseThresholdClassifier):
[[184 40]
[ 6 20]]
- # Multi-threshold classification using discretized scores
+ >>> # Multi-threshold classification using discretized scores
+ >>> import numpy as np
>>> base_clf = LogisticRegression(random_state=0)
>>> clf_multi = FixedThresholdClassifier(
... base_clf, threshold=[0.3, 0.6], labels=["Low", "Medium", "High"]
... ).fit(X_train, y_train)
>>> y_pred = clf_multi.predict(X_test)
- >>> import numpy as np
>>> labels, counts = np.unique(y_pred, return_counts=True)
>>> result = [(str(label), int(count)) for label, count in zip(labels, counts)]
>>> print(result)
@@ -327,7 +327,6 @@ class FixedThresholdClassifier(BaseThresholdClassifier):
High: 14
Low: 218
Medium: 18
- <BLANKLINE>
"""
_parameter_constraints: dict = { |
05bc405 to
f2ef469
Compare
|
Hello @adrinjalali . |
|
Yes, this requires an enhancement or a feature changelog entry. cc @glemaitre for a review. |
…-learn#30452 This commit adds support for multi-threshold classification to FixedThresholdClassifier, enabling discretization of continuous decision scores into multiple classes. The classifier now accepts a list of thresholds and optional labels for each bin. Includes: - Unit tests covering the new functionality - Documentation and usage example Co-authored-by: Pedro Lopes <[email protected]>
f2ef469 to
a29bef0
Compare
|
Hi @glemaitre , |
This commit adds support for multi-threshold classification to FixedThresholdClassifier, enabling discretization of continuous decision scores into multiple classes.
The classifier now accepts a list of thresholds and optional labels for each bin.
Includes:
Reference Issues/PRs
#30452
Multiple thresholds in FixedThresholdClassifier
What does this implement/fix? Explain your changes.
This PR extends FixedThresholdClassifier to support discretization of continuous decision scores into multiple classes, by allowing the threshold parameter to be a list of values. This enables users to bin decision function outputs (such as probabilities or scores) into more than two risk classes or categories, a common requirement in real-world applications such as credit scoring, risk modeling, and regulatory compliance.
Key additions: