ffselect

Fast feature subset selection library

This algorithm performs feature subset selection in O(n log n) or O(n) time

It may be useful for eliminating polynomial features with n equal to hundreds or thousands, where regular subset selection algorithms cannot perform in adequate time.

Usage

from ffselect.subset import MinSubsetSelection, FastSubsetSelection

MinSubsetSelection(data, target, fit_function, features, loss=True, interactive=True)
"""
Minimal feature subset selection algorithm
    data: Input data to pass to the fitting function
    target: Target parameter feature name
    fit_function: Callable function that fits the model and returns R^2/loss
    features: List of feature names
    loss: Set to true (by default) if the fitting function returns loss, R^2 otherwise
    interactive: Print output (default True)
    
    return: Tuple with the resulting R^2/loss and the list of features
"""


def FastSubsetSelection(data, target, fit_function, features, threshold = None, loss = True, interactive = True):
"""
Fast feature subset selection algorithm in linear time. May drop important features
    data: Input data to pass to the fitting function
    target: Target parameter feature name
    fit_function: Callable function that fits the model and returns R^2/loss
    features: List of feature names
    threshold: Minimal difference in loss/R^2 at which we drop the feature
    loss: Set to true (by default) if the fitting function returns loss, R^2 otherwise
    interactive: Print output (default True)
    
    return: Tuple with the resulting R^2/loss and the list of features
"""

Please view subset.ipynb for the complete example

Installation

pip3 install ffselect