R/stats.R

In stackcon/rngt: General data science and utilities package

#' Calculate Matthews correlation coefficient (MCC), F1, etc
#'
#' @param tmpcm confusion matrix that follows the default format returned by `caret::confusionMatrix`
#'
#' @return matthews correlation coef value
#' @export
#'
#' @examples
#' cm = matrix(c(20,5,
#'         10,15),
#'         nrow = 2,
#'         ncol = 2,
#'         byrow = TRUE,
#'         dimnames = list(c("pos", "neg"), c("pos", "neg")))
#' get_mcc(cm)  # 0.4082483
#' get_f1(cm)   # 0.7272727
#' get_cohens(cm) # 0.4
get_mcc = function(tmpcm){
	# confusion matrix where the,
	# ... rows indicate the prediction, and columns indicate the reference
	# ... first element of each row or column represents the "positive" class
	# ... second element of each row or column represents the "negative" class
	# tmpcm = [ TP, FP ]
	#			 [ FN, TN ]
	TP = tmpcm[1,1]
	TN = tmpcm[2,2]
	FP = tmpcm[1,2]
	FN = tmpcm[2,1]
	numer = (TP * TN) - (FP * FN)
	#denom =  (TP+FN) * (TP+FP) * (TN+FN) * (TN+FP)
	logdenom = log(TP+FN) + log(TP+FP) + log(TN+FN) + log(TN+FP)
	return( numer / exp(0.5 * logdenom) )
	#return( numer / sqrt(denom) )
}

#' @describeIn get_mcc calculate f1 score from confusion matrix
#' @export
get_f1 = function(tmpcm){
	TP = tmpcm[1,1]
	TN = tmpcm[2,2]
	FP = tmpcm[1,2]
	FN = tmpcm[2,1]
	(2*TP) / ( (2*TP) + FP + FN )
}

#' @describeIn get_mcc calculate Cohen's kappa coefficient from a confusion matrix
#' @export
get_cohens = function(tmpcm){
	# From [wikipedia](https://en.wikipedia.org/wiki/Cohen%27s_kappa):
	TP = tmpcm[1,1]
	TN = tmpcm[2,2]
	FP = tmpcm[1,2]
	FN = tmpcm[2,1]
	numer = 2 * (TP * TN - FP * FN)
	logdenom1 = log(TP+FP) + log(FP+TN)
	logdenom2 = log(TP+FN) + log(FN+TN)
	return( numer / (exp(logdenom1) + exp(logdenom2)) )
}