R/MboPlotFit-helpers.R

In PhilippScheller/VisBayesOpt: Visualizing model based optimization runs from package mlrMBO

# Helper functions for 'MboPlotFit'
# calculate R-squared based on 'summary-lm()' R-squared which is a robust version of the theoretically known R-squared.
RSQOverIterations = function(opt_path, names_x) {
  opt_path_x = opt_path[,names_x, drop = FALSE]
  y_hat = opt_path$mean
  y_eval = opt_path$y
  # p = ncol(opt_path_x) # might be needed for adjusted R2
  # n = nrow(opt_path_x) # might be needed for adjusted R2

  rss = sum((y_eval-y_hat)^2)
  tss = sum((y_eval-mean(y_eval))^2)
  if (tss == 0) {
    R2 = NA_real_
  } else {
    R2 = ifelse(1-(tss/(rss+tss)) < 0, 0, tss/(rss+tss)) # r2 numerical stable in accordante to r2 from summary.lm (in stats package)
  }
  return(R2)
}