R/simulate.R

In missCompare: Intuitive Missing Data Imputation Framework

#' @title Simulation of matrix with no missingness
#'
#' @description
#' \code{\link{simulate}} simulates a clean matrix with no missingness based on the original data structure
#' where all variables have the same mean and standard deviation and are normally distributed.
#'
#' @details
#' This function requires the metadata from the original dataframe and simulates a matrix with no missingness
#' with the same number of rows and columns and with the same or very similar correlation matrix as observed
#' in the original dataframe. When the correlation matrix is a non positive definitive matrix, the nearPD function
#' estimates the closest positive definitive matrix. Outputs from the function makes it easy to compare the original
#' correlation matrix with the nearPD correlation matrix. In the simulated matrix all variables have normal
#' distribution and fixed mean and standard deviation. This matrix will be subsequently used for spiking in missing
#' values and for the testing of various missing data imputation algorithms.
#'
#' @param rownum Number of rows (samples) in the original dataframe (Rows output from the \code{\link{get_data}} function)
#' @param colnum Number of rows (variables) in the original dataframe (Columns output from the \code{\link{get_data}} function)
#' @param cormat Correlation matrix of the original dataframe (Corr_matrix output from the \code{\link{get_data}} function)
#' @param meanval Desired mean value for the simulated variables, default = 0
#' @param sdval Desired standard deviation value for the simulated variables, default = 1
#'
#' @name simulate
#'
#' @return
#' \item{Simulated_matrix}{Simulated matrix with no missingness. The simulated matrix resembles the original dataframe in size and correlation structure, but has normally distributed variables with fixed means and SDs}
#' \item{Original_correlation_sample}{Sample of the original correlation structure (for comparison)}
#' \item{NearPD_correlation_sample}{Sample of the nearPD (nearest positive definitive matrix) correlation structure of the simulated matrix (for comparison)}
#'
#' @examples
#' cleaned <- clean(clindata_miss, missingness_coding = -9)
#' metadata <- get_data(cleaned)
#' simulated <- simulate(rownum = metadata$Rows, colnum = metadata$Columns,
#' cormat = metadata$Corr_matrix)
#'
#' @export


### FUNCTION
simulate <- function(rownum, colnum, cormat, meanval = 0, sdval = 1) {
    pd_corr_matrix <- Matrix::nearPD(cormat, keepDiag = TRUE, conv.tol = 1e-07, corr = TRUE)
    mu <- rep(meanval, colnum)
    stddev <- rep(sdval, colnum)
    covMat <- stddev %*% t(stddev) * pd_corr_matrix$mat
    X_hat <- MASS::mvrnorm(n = rownum, mu = mu, Sigma = covMat, empirical = TRUE)  # Simulated values
    if (colnum > 5)
        original_sample <- cormat[1:5, 1:5] else original_sample <- cormat[1:colnum, 1:colnum]
    if (colnum > 5)
        nearPD_sample <- stats::cor(X_hat)[1:5, 1:5] else nearPD_sample <- stats::cor(X_hat)[1:colnum, 1:colnum]

    rownames(X_hat) <- 1:nrow(X_hat)

    list(Simulated_matrix = X_hat, Original_correlation_sample = original_sample, NearPD_correlation_sample = nearPD_sample)
}