Caret code exmaples

Caret Models Examples.R

fitControl <- trainControl(
  method = 'LOOCV',
  savePredictions = TRUE,
  classProbs = TRUE,
  summaryFunction = twoClassSummary)

fitControlOversampling <- trainControl(
  method = 'LOOCV',
  savePredictions = FALSE,
  classProbs = TRUE,
  summaryFunction = twoClassSummary,
  sampling = "up"
)

# ===== LDA =====
# Center and scale at each CV iteration
lda.fit <- train(y=y, x=X, method="lda", preProcess=c("center", "scale"), trControl = fitControl)

# ===== GLM =====
glm.fit <- train(y=y, x=X, method="glm", preProcess=c("center", "scale"), trControl = trControl, family=binomial(link = "logit"))
  
# ===== rpart =====
Grid <- expand.grid(cp=seq(0, 0.2, 0.005))
fit.rpart <- train(y=y, x=X, method = 'rpart', trControl=fitControl, tuneGrid = Grid, na.action = na.pass)
# plot(fit.rpart)
# table(y, predict(fit.rpart, newdata = X, type = 'raw', na.action = na.pass))

# ===== rpart2 =====
Grid<-expand.grid(.maxdepth=2:20)
fit.rpart2 <- train(y=y, x=X, method = 'rpart2', trControl=fitControl, tuneGrid=Grid, na.action = na.pass)
# plot(fit.rpart2)
# table(y, predict(fit.rpart2, newdata = X, type = 'raw', na.action = na.pass))
  
# ===== C5.0 =====
Grid <- expand.grid(.trials = c(1:30), .model = c("tree", "rules"), .winnow = c(TRUE, FALSE))
fit.c5 <- train(y=y, x=X, method = 'C5.0', trControl=fitControl, tuneGrid=Grid, na.action = na.pass)
# plot(fit.c5)
# table(y, predict(fit.c5, newdata = X, type = 'raw', na.action = na.pass)

# ==== randomForest =====
Grid <- expand.grid(.mtry=seq(2, 30))
# OOB estimation can be used. ntree sets number of trees
fit.rf <- train(y=y, x=X, method = 'rf', trControl=trainControl(method = 'oob', savePredictions = FALSE), tuneGrid=Grid, ntree=2000)