Welcome to arfs’s documentation!

A package for performing All Relevant Feature Selection but not only that.

Documentation last change: October 21, 2026 at 14:14

User's guide

• Introduction
• Methods overview
  • Boruta
  • BoostARoota
  • Modifications to Boruta
  • BoostAGroota
  • GrootCV, a new method
  • MRmr
  • Lasso
  • References
• Documentation workflows
  • Runtime validation
  • Selection experiments
  • Opt-in MRMR row sampling
  • Build HTML locally
  • Example execution scope
  • Tree preprocessing tutorial revision
  • Sampling limits
• src
  • arfs package

Tutorials

• Preprocessing: check the partitions against known truth
  • Goal
  • Setup
  • A designed data-generating process
  • Check the individual learned partitions
    • Compare true effects with centered one-tree responses
  • Category renaming: a correctness counterexample
  • Composition, constants and deployment cases
  • Predictive comparison on untouched rows
  • Interpretation and next steps
• Preprocessing - Basic feature selection
  • Removing columns with many missing values
  • Removing columns with zero variance
  • Removing columns with high cardinality
  • Removing columns highly correlated
  • Removing columns with zero or low predictive power
  • All at once - Sklearn Pipeline
• Association and Feature Selection
  • Default association functions
  • Association series and matrix
  • User defined association functions
    • Example
• Link to feature selection
• ARFS - classification
  • Simple Usage
  • Leshy
  • BoostAGroota
  • GrootCV
  • ARFS in sklearn pipelines
  • Testing and comparing Leshy, GrootCV and BoostAGroota
    • Testing Leshy
    • Testing GrootCV
    • Testing BoostAGroota
    • comparing to BorutaShap
• ARFS - regression
  • Simple Usage
  • Leshy (Boruta evolution)
  • BoostAGroota
  • GrootCV
  • ARFS in sklearn pipelines
  • Testing and comparing Leshy, GrootCV and BoostAGroota
    • Testing Leshy
    • Testing GrootCV
    • Testing BoostAGroota
• ARFS - user defined cross validation scheme - Time series applicaition
• ARFS - How to use with large data?
  • Sampling data
  • Cleaning the data
  • Random sampling
  • Sampling by clustering the rows
  • Sampling by anomaly-score similarity
  • Impact on the feature selection
• ARFS - Using GPU
  • GrootCV on GPU
  • Leshy on GPU
• ARFS - fasttreeshap vs shap
  • GrootCV - fastshap vs shap
    • Fastshap enable
    • Fastshap disable
• ARFS - Customizing GrootCV
  • Number of processes
  • Regularization
• ARFS vs Boruta and BorutaShap
  • Comparison
    • BorutaPy
    • Leshy
    • Same Results?
    • BorutaShap with native importance
    • Comparison with Boruta
• ARFS - Non Normal loss and Sample Weight
  • Leshy
  • BoostAGRoota
  • MRmr
• MRMR: selection before and after supervised partitioning
  • Local regression example
  • Local classification comparison
  • Interpretation
• ARFS vs MRmr
  • MRmr
  • ARFS - GrootCV
• Lasso selection: ordinary factors versus supervised partitions
  • Linear model
  • A toy example using the selector in a pipeline
  • Poisson counts observed over synthetic exposure
  • Weighted categorical regression: a fair comparison
    • Formula-based designs
    • Interpretation
• ARFS with Categorical predictors
  • Simple Usage
  • GrootCV
  • BoostAGroota
  • Leshy
• ARFS - Is collinearity harmul?
  • Simple Usage
  • Sklearn permutation importance
  • Leshy with permutation importance
  • Does SHAP feature importance suffer from the collinearity?
    • With all the predictors
    • With only the filtered predictors
  • BoostAGroota
    • All the predictors
    • Filtered predictors
  • GrootCV
    • All the predictors
    • Filtered predictors

Indices and tables

• Index

• Module Index

• Search Page