impute_matrix: Impute missing values in given matrix

In Columbia-PRIME/pcpr: Principal Component Pursuit for Environmental Epidemiology

Impute missing values in given matrix

Description

impute_matrix() imputes the missing NA values in a given matrix using a given imputation_scheme.

Usage

impute_matrix(D, imputation_scheme)

Arguments

D

The input data matrix.

imputation_scheme

The values to replace missing NA values in D with. Can be either: A scalar numeric, indicating all NA values should be imputed with the same scalar numeric value; A vector of length ncol(D), signifying column-specific imputation, where each entry in the imputation_scheme vector corresponds to the imputation value for each column in D; or A matrix of dimension dim(D), indicating an observation-specific imputation scheme, where each entry in the imputation_scheme matrix corresponds to the imputation value for each entry in D.

Value

The imputed matrix.

See Also

sim_na(), sim_lod(), sim_data()

Examples

#### ------------Imputation with a scalar------------####
# simulate a small 5x5 mixture
D <- sim_data(5, 5)$D
# corrupt the mixture with 40% missing observations
D_tilde <- sim_na(D, 0.4)$D_tilde
D_tilde
# impute missing values with 0
impute_matrix(D_tilde, 0)
# impute missing values with -1
impute_matrix(D_tilde, -1)

#### ------------Imputation with a vector------------####
# impute missing values with the column-mean
impute_matrix(D_tilde, apply(D_tilde, 2, mean, na.rm = TRUE))
# impute missing values with the column-min
impute_matrix(D_tilde, apply(D_tilde, 2, min, na.rm = TRUE))

#### ------------Imputation with a matrix------------####
# impute missing values with random Gaussian noise
noise <- matrix(rnorm(prod(dim(D_tilde))), nrow(D_tilde), ncol(D_tilde))
impute_matrix(D_tilde, noise)

#### ------------Imputation with LOD/sqrt(2)------------####
D <- sim_data(5, 5)$D
lod_info <- sim_lod(D, q = 0.2)
D_tilde <- lod_info$D_tilde
D_tilde
lod <- lod_info$lod
impute_matrix(D_tilde, lod / sqrt(2))

Columbia-PRIME/pcpr documentation built on April 14, 2025, 8:33 a.m.