feature_combinations: Define feature combinations, and fetch additional information...

In shapr: Prediction Explanation with Dependence-Aware Shapley Values

Define feature combinations, and fetch additional information about each unique combination

Description

Define feature combinations, and fetch additional information about each unique combination

Usage

feature_combinations(
  m,
  exact = TRUE,
  n_combinations = 200,
  weight_zero_m = 10^6
)

Arguments

m	Positive integer. Total number of features.
exact	Logical. If TRUE all 2^m combinations are generated, otherwise a subsample of the combinations is used.
n_combinations	Positive integer. Note that if exact = TRUE, n_combinations is ignored. However, if m > 12 you'll need to add a positive integer value for n_combinations.
weight_zero_m	Numeric. The value to use as a replacement for infinite combination weights when doing numerical operations.

Value

A data.table that contains the following columns:

id_combination

Positive integer. Represents a unique key for each combination. Note that the table is sorted by id_combination, so that is always equal to x[["id_combination"]] = 1:nrow(x).

features

List. Each item of the list is an integer vector where features[[i]] represents the indices of the features included in combination i. Note that all the items are sorted such that features[[i]] == sort(features[[i]]) is always true.

n_features

Vector of positive integers. n_features[i] equals the number of features in combination i, i.e. n_features[i] = length(features[[i]]).

.

N

Positive integer. The number of unique ways to sample n_features[i] features from m different features, without replacement.

Author(s)

Nikolai Sellereite, Martin Jullum

Examples

# All combinations
x <- feature_combinations(m = 3)
nrow(x) # Equals 2^3 = 8

# Subsample of combinations
x <- feature_combinations(exact = FALSE, m = 10, n_combinations = 1e2)

shapr documentation built on May 4, 2023, 5:10 p.m.