weights_mpsi <- function(df)
{
  criteria_direction <- df[1, 2:ncol(df)]
  criterias <- colnames(df)[2:ncol(df)]
  matrix_decis <- as.data.frame(df[3:nrow(df), 2:ncol(df)])
  colnames(matrix_decis) <- criterias
  matrix_decis <- matrix_decis %>% mutate_all(as.numeric)
  matrix_decis_norm <- data.frame(matrix(NA, ncol = ncol(matrix_decis),
                                         nrow = nrow(matrix_decis)))

  for(i in 1:ncol(matrix_decis_norm)) {
    if(criteria_direction[i] == "MAX") {
      matrix_decis_norm[,i] <- matrix_decis[,i] / max(matrix_decis[,i])
    } else {
      matrix_decis_norm[,i] <- min(matrix_decis[,i]) / matrix_decis[,i]
    }
  }
  pv <- rep(0, ncol(matrix_decis_norm))
  for(j in 1:ncol(matrix_decis_norm)) {
    for(i in 1:nrow(matrix_decis_norm)) {
      pv[j] <- pv[j] + ((matrix_decis_norm[i,j]) -
                          mean(matrix_decis_norm[,j]))^2
    }
  }
  weights_MPSI <- pv / sum(pv)
  weights_MPSI <- as.data.frame(t(weights_MPSI))
  colnames(weights_MPSI) <- criterias
  return(weights_MPSI)
}
#
# pesos_mpsi <- weights_mpsi(df)
# ordem_psi <- rank_psi(df, pesos_mpsi)
# ordem_cocoso <- rank_cocoso(df, weights_cocoso = pesos_mpsi, alpha = 0.5)
# ordem_topsis <- rank_topsis(df, weights_topsis = pesos_mpsi)