library(stats19)
library(tidyverse)
library(MASS)
library(pixmap)

calibra <- function(dataset = grep("normal|happy|sad|sleepy|surprised|wink",
                                   list.files("data_atividade2/"), value=TRUE)) {
  set.seed(202227)
  data <- matrix(ncol = 20691, nrow = 90)
  c <- 1
  for(i in dataset){#monta matriz do poligono geral de um rosto das pessoas
    path <- paste("data_atividade2", i, sep="/")
    img <- read.pnm(file=path, cellres=1)
    vetor_rosto <- as.numeric(img@grey[40:210, 40:160])
    data[c,] <- vetor_rosto
    c <- c + 1
  }
  data <- as.data.frame(data)
  data$resp <- as.factor(rep(c("happy", "normal", "sad", "sleepy", "surprised","wink"), 15))

  id <- c(1:nrow(data))
  id_train <- sample(id, 0.8*nrow(data))
  train <- data %>% slice(id_train)
  test <- data %>% slice(-id_train)

  pc <- prcomp(train[-20692], scale=TRUE, retx=FALSE)
  train.pc <- as.data.frame(predict(pc, train))
  test.pc <- as.data.frame(predict(pc, test))
  var.pc.prop <- (pc$sdev^2)/sum(pc$sdev^2)
  cum.var <- cumsum(var.pc.prop)
  ncomp <- which.max(cum.var > 0.95)

  train.pc <- train.pc[1:ncomp]
  train.pc$resp <- train$resp
  test.pc <- test.pc[1:ncomp]
  test.pc$resp <- test$resp
  lda.fit <- lda(resp ~ ., data = train.pc)

  final <- list(lda.fit, pc, ncomp)
  return(final)
}

prediz <- function(modelo=calibra()[[1]], teste){
  img <- matrix(ncol=20691)
  vetor_rosto <- as.numeric(teste[40:210, 40:160])
  img <- rbind(img, vetor_rosto) %>% as.data.frame()
  img <- img[2,]
  teste_proc <- as.data.frame(predict(calibra()[[2]], img))
  teste_proc <- teste_proc[1:calibra()[[3]]]
  teste_pred <- predict(modelo, teste_proc)
  return(teste_pred$class)





  #teste.pc <- as.data.frame(predict(calibra()[[2]], teste))
  #teste.pc <- teste.pc[1:calibra()[[3]]]
  #teste.pred <- predict(modelo, teste.pc)
  #return(teste.pred$class)
}