getBRMModelPrediction.R

library(tidyverse)
library(caret)

#' Predict values from an ordinal brm regression model.
#' 
#' @param model the fitted model.
#' @param data a datafram with all explanatory variables and labels.
#' @return a dataframe with a probability value for each labels, for each item, and true value.
#' @examples
#' add(1, 1)
#' add(10, 1)
getPredictions <- function(model ,data, labeled_column) {
  output = predict(
    model,
    newdata = data,
    re_formula = NA,
  )
  labels = trimws(gsub('P\\(Y =', "", gsub('\\)','',colnames(output))))
  output = as.data.frame(output)
  colnames(output)=labels
  output = output %>% rownames_to_column() %>% gather(predict_label, value, -rowname) %>% group_by(rowname) %>%  filter(rank(-value) == 1)
  data = data %>% rownames_to_column() %>% select (rowname,all_of(labeled_column))
  output = output  %>% inner_join(data, by = "rowname",na_matches="never")
  output$is_ok = output$predict_label == output[,c(labeled_column)]
  confusion_matrix <- confusionMatrix(output$predict_label,  output[,c(labeled_column)
  print(confusion_matrix)
  return(output)
  }

• filter(rank(-value) == 1) not get a value where there is several maximum value