Yioop_V9.5_Source_Code_Documentation

ChiSquaredFeatureSelection extends FeatureSelection
in package

A subclass of FeatureSelection that implements chi-squared feature selection.

This feature selection method scores each feature according to its informativeness, then selects the top N most informative features, where N is a run-time parameter.

Tags
author

Shawn Tice

Table of Contents

$max  : int
The maximum number of features to select, a runtime parameter.
__construct()  : mixed
Sets any passed runtime parameters.
buildMap()  : array<string|int, mixed>
Constructs a map from old feature indices to new ones according to a max-heap of the most informative features. Always keep feature index 0, which is used as an intercept term.
select()  : array<string|int, mixed>
Uses the chi-squared feature selection algorithm to rank features by informativeness, and return a map from old feature indices to new ones.

Properties

Methods

__construct()

Sets any passed runtime parameters.

public __construct([array<string|int, mixed> $parameters = [] ]) : mixed
Parameters
$parameters : array<string|int, mixed> = []

optional associative array of parameters to replace the default ones with

Return values
mixed

buildMap()

Constructs a map from old feature indices to new ones according to a max-heap of the most informative features. Always keep feature index 0, which is used as an intercept term.

public buildMap(object $selected) : array<string|int, mixed>
Parameters
$selected : object

max heap containing entries ordered by informativeness and feature index.

Return values
array<string|int, mixed>

associative array mapping a subset of the original feature indices to the new indices

select()

Uses the chi-squared feature selection algorithm to rank features by informativeness, and return a map from old feature indices to new ones.

public select(object $features) : array<string|int, mixed>
Parameters
$features : object

full feature set

Return values
array<string|int, mixed>

associative array mapping a subset of the original feature indices to new indices


        

Search results