Recipes

library(tidyverse)
library(recipes)
library(sjPlot)
library(caret)
library(doParallel)
library(GGally)
library(Boruta)

data("okc")
str(okc)

##tratar fatores NA
okc <- okc %>%
  mutate_at(c('diet','location'), funs(ifelse(is.na(.),'indisponible',.)))


rec <- recipe(Class ~ . , data = okc) %>%
  step_other(all_nominal(), -all_outcomes(), threshold = 0.05, other = 'other values') %>%
  step_holiday(date) %>%
  step_dummy(all_nominal(), -all_outcomes()) %>%
  prep()

okc_treat <- bake(rec, okc)

ok_treatX <- bake(rec, okc, composition = 'matrix', all_predictors(), -date)
ok_treatY <- okc$Class

model_01 <- glm(Class ~ age + diet + height + location, data = okc_treat,
                family = 'binomial')
summary(model_01)
tab_model(model_01)


############################################
data("segmentationData")
str(segmentationData)
dim(segmentationData)

seg_train <- segmentationData %>% filter(Case == 'Train') %>% select(-Case)
seg_test <- segmentationData %>% filter(Case == 'Test') %>% select(-Case)

rec <- recipe(Cell ~ . , data = seg_train) 

rec <- rec %>%
  step_YeoJohnson(all_numeric()) %>%
  step_center(all_numeric()) %>%
  step_scale(all_numeric()) %>%
  step_pca(all_numeric(), -all_outcomes())

summary(rec)
preparo <- prep(rec, seg_train, retain = TRUE)

seg_train <- bake(preparo, seg_train, -Cell)
seg_test <- bake(preparo, seg_test, -Cell)

##Matrix interface
seg_trainX <- bake(preparo, seg_train, -Cell, -Class, composition = 'matrix')
seg_trainY <- seg_train$Class
seg_testX <- bake(preparo, seg_test, -Cell, -Class, composition = 'matrix')
seg_testY <- seg_test$Class


## ggplot
ggplot(seg_train, aes(PC1, PC2, color = Class))+
  geom_point()


##Caret
set.seed(1234)
control <- trainControl(
  method = 'cv',
  number = 10,
  summaryFunction = twoClassSummary,
  classProbs = TRUE)

##SVM
set.seed(1234)
model_svm <- train(rec, 
                   data = seg_train,
                   method = "svmRadial",
                   metric = 'ROC',
                   tuneLenght = 10,
                   trControl = control)

predito <- predict(model_svm, seg_test)
confusionMatrix(predito, seg_test$Class)


##SVM with matrix
set.seed(1234)
model_svm <- train(x = seg_trainX, y = seg_trainY,
                   method = "svmRadial",
                   metric = 'ROC',
                   tuneLenght = 10,
                   trControl = control)

predito <- predict(model_svm, seg_testX)
confusionMatrix(predito, seg_testY)



##gbm with matrix
set.seed(1234)
model_lgbm <- train(x = seg_trainX, y = seg_trainY,
                   method = "gbm",
                   metric = 'ROC',
                   trControl = control)

model_lgbm
predito <- predict(model_lgbm, seg_testX)
confusionMatrix(predito, seg_testY)


##nnet with matrix
set.seed(1234)
model_net <- train(x = seg_trainX, y = seg_trainY,
                    method = "nnet",
                    metric = 'ROC',
                    trControl = control)

predito <- predict(model_net, seg_testX)
confusionMatrix(predito, seg_testY)


################data Credit_data #############################33
data("credit_data")
str(credit_data)
summary(credit_data)


##Substitui factores na
credit <- credit_data %>%
  mutate(Home = as.character(Home),
         Marital = as.character(Marital),
         Job = as.character(Job)) %>%
  mutate(Home = ifelse(is.na(Home),'indisponible',Home),
         Marital = ifelse(is.na(Marital),'indisponible',Marital),
         Job = ifelse(is.na(Job),'indisponible',Job) )

##Some plots
ggplot(credit, aes(Expenses, fill = Status))+
  geom_histogram()

ggplot(credit, aes(Status, log(Income), fill = Status))+
  geom_boxplot()

ggplot(credit, aes(Status, log(Seniority), fill = Status))+
  geom_boxplot()

ggplot(credit, aes(Status, Age, fill = Status))+
  geom_boxplot()

##intereção das princiapis variaveis
# Income, Seniority, Amount, Price, Age by Status

ggpairs(credit[c('Income','Seniority','Amount','Price','Age','Status')], aes(color = Status))

##Separaret File
index <- createDataPartition(credit$Status, p = 0.75, list = FALSE)
credit_train <- credit[index, ]
credit_test <- credit[-index, ]

#recipe
rec <- recipe(Status ~ . , data = credit_train) %>%
  step_YeoJohnson(all_numeric()) %>%
  step_center(all_numeric()) %>%
  step_scale(all_numeric()) %>%
  step_medianimpute(all_numeric()) %>%
  step_corr(all_numeric(), threshold = 0.9) %>%
  step_other(-all_numeric(), -all_outcomes(), threshold = 0.05, other = 'other level') %>%
  step_dummy(-all_numeric(), -all_outcomes()) %>%
  step_nzv(all_predictors())

preparo <- prep(rec, credit_train, retain = TRUE)

credit_trainX <- bake(preparo, credit_train, -all_outcomes(), composition = 'matrix')
credit_trainY <- credit_train$Status
credit_testX <- bake(preparo, credit_test, -all_outcomes(), composition = 'matrix')
credit_testY <- credit_test$Status


##Boruta
feature <- Boruta(x = credit_trainX, y = credit_trainY, doTrace = 2)
plot(feature)
feature$finalDecision
getConfirmedFormula(feature)

## DoParalell
cl <- makePSOCKcluster(5)
registerDoParallel(cl)

##Visualização
rec2 <- rec %>%
  step_kpca(all_predictors()) %>%
  prep()

credit_pca_train <- bake(rec2, credit_train)
ggplot(credit_pca_train, aes(kPC1, kPC2, color = Status))+
  geom_point()

##Caret
set.seed(1234)
control <- trainControl(
  method = 'cv',
  number = 10,
  summaryFunction = twoClassSummary,
  classProbs = TRUE)

##SVM
set.seed(1234)
model_svm <- train(x = credit_trainX,
                   y = credit_trainY,
                   method = "svmRadial",
                   metric = 'ROC',
                   tuneLenght = 10,
                   trControl = control)

model_svm
plot(varImp(model_svm))
predito <- predict(model_svm, credit_testX)
confusionMatrix(predito, credit_testY)

##Rf
set.seed(1234)
model_rf <- train(x = credit_trainX,
                   y = credit_trainY,
                   method = "rf",
                   metric = 'ROC',
                   tuneLenght = 10,
                   trControl = control)

model_rf
plot(varImp(model_rf))
predito <- predict(model_rf, credit_testX)
confusionMatrix(predito, credit_testY)


##gbm 
set.seed(1234)
model_gbm <- train(x = credit_trainX,
                  y = credit_trainY,
                  method = "gbm",
                  metric = 'ROC',
                  trControl = control)

model_gbm
plot(varImp(model_gbm))
predito <- predict(model_gbm, credit_testX)
confusionMatrix(predito, credit_testY)

##H20
library(h2o)
h2o.init()

h2o_train <- bake(preparo, credit_train)
h2o_test <- bake(preparo, credit_test)

h2o_train <- as.h2o(h2o_train)
h2o_test <- as.h2o(h2o_test)

# Model GBM
credit.gbm <- h2o.gbm(x = 2:18, y = 1, training_frame = h2o_train, ntrees = 10,
                   max_depth = 4, min_rows = 2, learn_rate = 0.2,
                   nfolds = 10)

h2o.performance(credit.gbm, h2o_test)


# Model Rf
credit.rf <- h2o.randomForest(x = 2:18, y = 1, training_frame = h2o_train,
                           nfolds = 10)

h2o.performance(credit.rf, h2o_test)