CalfCV Examples Gallery#

This gallery provides practitioner-focused examples demonstrating how to leverage CalfCV for automated hyperparameter tuning, dynamic noise suppression, and highly interpretable feature selection.

Together, these scripts illustrate how CalfCV trades upfront computational time for robust generalization, yielding simple integer-weighted models that match the predictive power of continuous optimization techniques.

Gallery Highlights#

Dynamic Noise Suppression: Demonstrates how CalfCV dynamically prunes pure noise features without requiring a hardcoded feature count target (k), protecting downstream classifiers from overfitting. (See: CALF as a Supervised Feature Selection Preprocessor)
Clinical Interpretability vs. Continuous Weights: Validates the estimator on real-world medical data. Compares CalfCV against L1 (Lasso) and RFE, showing how discrete ±1 weights create a highly interpretable aggregate scorecard with competitive ROC-AUC. (See: Feature Selection Sparsity: CALF vs. L1 & RFE)
Runtime vs. Performance Benchmarking: Quantifies the computational cost of the internal grid search. Illustrates that the modest increase in fit time buys optimal predictive generalization and extreme model sparsity. (See: Runtime vs. Classifier Performance Trade-offs)
Decision Calibration & Probabilities: Proves that despite utilizing discrete integer weights, CalfCV produces calibrated probability estimates via a sigmoid transformation, allowing for standard precision-recall threshold tuning. (See: Decision Threshold Calibration)

Decision Threshold Calibration: Precision, Recall, and F1 Trade-offs

Cumulative AUC by Feature: Forward Selection Trajectory

Multilabel Text Document Classification with CALF One-vs-Rest

Feature Selection Sparsity: CALF vs. L1 & RFE

Sentiment Analysis of High-Dimensional IMDB Reviews

Runtime vs. Classifier Performance Trade-offs

CALF as a Supervised Feature Selection Preprocessor

Gallery generated by Sphinx-Gallery