Scikit-learn Interface¶

The wrapper module provides a scikit-learn compatible interface for threshold optimization.

ThresholdOptimizer Class¶

Usage Examples¶

Basic Usage¶

from optimal_cutoffs import ThresholdOptimizer
from sklearn.ensemble import RandomForestClassifier
from sklearn.model_selection import train_test_split

# Train your classifier
clf = RandomForestClassifier()
clf.fit(X_train, y_train)

# Get probabilities
y_prob_train = clf.predict_proba(X_train)[:, 1]  # Binary case
y_prob_test = clf.predict_proba(X_test)[:, 1]

# Optimize threshold
optimizer = ThresholdOptimizer(metric='f1', method='auto')
optimizer.fit(y_train, y_prob_train)

# Make predictions
y_pred = optimizer.predict(y_prob_test)

Multiclass Usage¶

# For multiclass problems, pass full probability matrix
y_prob_train = clf.predict_proba(X_train)  # Shape: (n_samples, n_classes)
y_prob_test = clf.predict_proba(X_test)

# Optimizer automatically detects multiclass
optimizer = ThresholdOptimizer(metric='f1')
optimizer.fit(y_train, y_prob_train)  # y_train has integer class labels

# Returns class predictions
y_pred = optimizer.predict(y_prob_test)

Pipeline Integration¶

from sklearn.pipeline import Pipeline
from sklearn.preprocessing import StandardScaler

# Create pipeline with threshold optimization
pipeline = Pipeline([
    ('scaler', StandardScaler()),
    ('classifier', RandomForestClassifier()),
    ('threshold', ThresholdOptimizer(metric='f1'))
])

# Note: This requires custom pipeline steps for probability extraction
# See advanced examples for full implementation

Attributes¶

After fitting, the ThresholdOptimizer instance has several useful attributes:

threshold_: The optimized threshold(s)
score_: The metric value achieved at the optimal threshold
n_classes_: Number of classes detected (1 for binary, >1 for multiclass)
classes_: Array of unique class labels