R/viz_pcacm.R

In EmilHvitfeldt/horus: Visual Tools To Help Machine Learning model Selection

#' PCA component measures
#'
#' @param x a data.frame
#' @param n_pca Number of Principle components to show
#' @param n_var Number of Variables to show
#'
#' @return a ggplot2 object
#' @export
#'
#' @examples
#' viz_pcacm(mtcars)
#' viz_pcacm(USArrests)
#' viz_pcacm(beaver1)
#'
#' viz_pcacm(mtcars, n_pca = 4)
#' viz_pcacm(mtcars, n_var = 4)
#' viz_pcacm(mtcars, n_pca = 2, n_var = 6)
viz_pcacm <- function(x, n_pca = 10, n_var = 10) {
  tidy_rotation <- as.data.frame(prcomp(x, rank = n_pca)$rotation) %>%
    rownames_to_column("var") %>%
    pivot_longer(-var) %>%
    filter(name %in% paste0("PC", seq_len(n_pca)))

  order <- tidy_rotation %>%
    group_by(var) %>%
    summarize(contribution = sum(abs(value) * 1 / parse_number(name))) %>%
    arrange(desc(contribution)) %>%
    slice(seq_len(n_var)) %>%
    pull(var)

  tidy_rotation %>%
    filter(var %in% order) %>%
    mutate(var = fct_relevel(var, order),
           name = fct_relevel(name, paste0("PC", seq_len(n_pca))),
           name = fct_rev(name)) %>%
    ggplot(aes(var, name, fill = value)) +
    geom_raster() +
    scale_fill_gradient2(
      low = "purple",
      mid = "grey90",
      high = "orange"
    ) +
    labs(
      y = "Principle Component",
      x = "Variable"
    ) +
    theme_minimal()
}