library(tidyverse)
library(caret)
library(here)
library(inspectdf)
library(glmnet)
library(ROSE)
library(pROC)
library(ppls)

rm(seqs)
load(here("notebooks/data/nback_seqs.Rd"))
seqs <- seqs %>% drop_na(rt, correct, tl,sl)

f <- correct ~ n + t + stimulus_type
f <- correct ~ n + tl + vl + sl + s + stimulus_type

# predictors selected with stepAIC
f <- correct ~ n + tl + t + l + stimulus_type
set.seed(654321)

train.indices <- createDataPartition(seqs[[toString(f[[2]])]], p = .8, list =FALSE)

seqs.train.balanced <- seqs[train.indices,]
#seqs.train <- ROSE(f, data = seqs.train.balanced, N=100)$data

seqs.train.x <- model.matrix(f, seqs.train)[,-1]
seqs.train.y <- seqs.train[[toString(f[[2]])]]

seqs.test  <- seqs[-train.indices,]
seqs.test.x <-  model.matrix(f, seqs.test)[,-1]
seqs.test.observed_y <- seqs.test[[toString(f[[2]])]]


set.seed(10)

ctrl <- trainControl(
  method = "cv",
  number = 10,
  classProbs = T,
  summaryFunction = twoClassSummary,
  sampling = "down"
)

grid <- expand.grid(
  alpha = seq(0,1,length=10),
  lambda = seq(0.0001, 0.1, length=10)
)

model <- train(seqs.train.x, seqs.train.y,
               method = "glmnet",
               preProcess = c("nzv","center","scale"), # or c("knnImpute"),
               tuneGrid = grid,
               metric = "ROC",
               trControl = ctrl)

seqs.test.y_prob <- predict(model, seqs.test.x, type="prob")

model
plot(varImp(model, useModel = F))
plot(model)
max(model$results$ROC)


roc(seqs.test.observed_y,
    seqs.test.y_prob$YES,
    legacy.axes=T,
    plot = T,
    lwd=2,
    col="black",
    print.auc=T,
    percent = T,
    print.auc.y = 40,
    print.auc.x = 55,
    lty = 1,
    of = "se",
    boot.n = 2000,
    ci = T)



# PPLS
#penalized.pls.cv(seqs.train.y, seqs.train.x, kernel = T, scale=T)