Inclass p3 push to repo

                  trControl = ctrl ,

                  preProcess = c("center" ,"scale") ,

                  tuneLength = 10)

# Model evaluation

XX_valid <- subset(validSet,select = -c(file,indicator))

pcs_new <- predict(pcs,newdata = XX_valid)

pred_age <- predict(knnFit,newdata=pcs_new)

prd_outcome_comparison <- cbind(as.numeric(pred_age)-1,validSet$indicator)

# Generate the probabilities of default

# Clear workspace and graphs

if(!is.null(dev.list())) dev.off()

rm(list = ls())

library("caret")

library("data.table")

load(file="01_data/ML_data.Rdata")

train_preProcess <- preProcess(trainSet[,-c("indicator","file")],method=c("center","scale"))

valid_preProcess <- preProcess(validSet[,-c("indicator","file")],method=c("center","scale"))

trainSet_stdz <- predict(train_preProcess,trainSet[,-c("file")])

validSet_stdz <- predict(valid_preProcess,validSet[,-c("file")])

ctrl <- trainControl(method="cv",number=10,verbose=TRUE)

tuneGrid <- expand.grid( alpha= 1, lambda = seq(0.1,1,0.1))

lasso_fit <- train(y = as.factor(trainSet_stdz$indicator),

      x = trainSet_stdz[,-c("indicator")],

      method="glmnet",

      metric="Accuracy",

      trControl = ctrl,

      tuneGrid = tuneGrid)

pred <- predict(lasso_fit,validSet_stdz)

# Alternative:

#predictions_lasso <- lasso_fit %>% predict(validSet) 

bestTuneModelCoefs <- as.data.frame.matrix(coef(lasso_fit$finalModel, lasso_fit$bestTune$lambda))

bestTuneModelCoefs$mz_vals <- as.numeric(rownames(bestTuneModelCoefs))

bestTuneModelCoefs[head(order(bestTuneModelCoefs$s1,decreasing=TRUE),10),]