observation_impute_cpp: Get imputed data
In shapr: Prediction Explanation with Dependence-Aware Shapley Values

observation_impute_cpp

R Documentation

Get imputed data

Description

Get imputed data

Usage

observation_impute_cpp(index_xtrain, index_s, x_train, x_explain, S)

Arguments

`index_xtrain`	Positive integer. Represents a sequence of row indices from `x_train`, i.e. `min(index_xtrain) >= 1` and `max(index_xtrain) <= nrow(x_train)`.
`index_s`	Positive integer. Represents a sequence of row indices from `S`, i.e. `min(index_s) >= 1` and `max(index_s) <= nrow(S)`.
`x_train`	Matrix. Contains the training data.
`x_explain`	Matrix with 1 row. Contains the features of the observation for a single prediction.
`S`	arma::mat. Matrix of dimension (`n_coalitions`, `n_features`) containing binary representations of the used coalitions. S cannot contain the empty or grand coalition, i.e., a row containing only zeros or ones. This is not a problem internally in shapr as the empty and grand coalitions are treated differently.

Details

S(i, j) = 1 if and only if feature j is present in feature combination i, otherwise S(i, j) = 0. I.e. if m = 3, there are 2^3 = 8 unique ways to combine the features. In this case dim(S) = c(8, 3). Let's call the features x1, x2, x3 and take a closer look at the combination represented by s = c(x1, x2). If this combination is represented by the second row, the following is true: S[2, 1:3] = c(1, 1, 0).

The returned object, X, is a numeric matrix where dim(X) = c(length(index_xtrain), ncol(x_train)). If feature j is present in the k-th observation, that is S[index_[k], j] == 1, X[k, j] = x_explain[1, j]. Otherwise X[k, j] = x_train[index_xtrain[k], j].