Philipp-Sc / importance

Permutation feature importance

Permutation feature importance

Feature selection

Feature importance is often used for variable selection.

Compute the relative importance of input variables of trained predictive models using feature shuffling

When called, the importance function shuffles each feature n times and computes the difference between the base score (calculated with original features X and target variable y) and permuted data. Intuitively that measures how the performance of a model decreases when we "remove" the feature.

• More info about the method: Permutation Feature Importance
• Permutation importance can be biased if features are highly correlated (Hooker, Mentch 2019)

Usage

importance = { git="https://github.com/philipp-sc/importance.git" }

    use importance::*;
    use importance::score::*;

    struct MockModel;

    impl Model for MockModel {
        fn predict(&self, x: &Vec<Vec<f64>>) -> Vec<f64> {
            x.iter().map(|x| x.iter().sum()).collect() // Mock predictions
        }
    }
 
    fn main() {
        let model = MockModel;
        let x = vec![vec![1.0, 0.0, 3.0], vec![4.0, 0.0, 6.0], vec![7.0, 0.0, 9.0]];
        let y = vec![4.0, 10.0, 16.0];

        let opts = Opts {
            verbose: true,
            kind: Some(ScoreKind::Smape),
            n: Some(100),
            only_means: true,
            scale: true,
        };

        let importances = importance(&model, x, y, opts);
        println!("Importances: {:?}", importances);
    }

API

importance(model, X, y, options)

• model - trained model with predict method
• X - 2D array of features
• y - 1D array of target variables

Options:

• kind - scoring function (mse, mae, rmse, smape, acc, ce (cross-entropy)
• n - number of times each feature is shuffled.
• only_means - if true returns only average importance
• verbose - if true throws some info into console

This is a rust port of https://github.com/zemlyansky/importance