Machine Learning With H2O Part 3 Solutions

###############
#             #
# Exercise 1  #
#             #
###############

setwd("H20/")
library(h2o)
cluster.h <- h2o.init()
bank_data <- h2o.importFile("data\bank.csv")
response = "y"

predictors <- c("age","job","marital","education","default","balance","housing","loan")


splits <- h2o.splitFrame(bank_data,c(0.8,0.1))
train <- splits[[1]]
valid <- splits[[2]]
test <- splits[[3]]

gbm.m <- h2o.gbm(x, y, train, nfolds = 10, model_id = "GBM_defaults",seed = 1042)

b <- h2o.gbm(x = predictors, y = response, training_frame = train,
                   validation_frame = valid, balance_classes = TRUE, seed = 1234,nfolds = 5,keep_cross_validation_predictions = TRUE,ntrees = 10)

h2o.auc(h2o.performance(b,newdata = valid))
h2o.auc(h2o.performance(b,newdata = test))

###############
#             #
# Exercise 2  #
#             #
###############
class_sampling_factor = c(0.2,1.0)
b.1 <- h2o.gbm(x = predictors, y = response, training_frame = train,
             validation_frame = valid, balance_classes = TRUE,
             seed = 1234,nfolds = 10,ntrees = 15,class_sampling_factors = class_sampling_factor,
              fold_assignment = "Stratified",distribution = "bernoulli",
             stopping_metric = "AUC",stopping_tolerance = 1e-2,stopping_rounds = 4
             )


b.2<- h2o.gbm(x = predictors, y = response, training_frame = train,
               validation_frame = valid,
               balance_classes = TRUE, seed = 1234,nfolds = 10,
               ntrees = 10,class_sampling_factors = c(1.0,0.2),
              fold_assignment = "Stratified",distribution = "bernoulli",
              stopping_metric = "AUC",stopping_tolerance = 1e-2,stopping_rounds = 4)


h2o.auc(h2o.performance(b.1,newdata = valid))
h2o.auc(h2o.performance(b.1,newdata = test))

h2o.auc(h2o.performance(b.2,newdata = valid))
h2o.auc(h2o.performance(b.2,newdata = test))



###############
#             #
# Exercise 3  #
#             #
###############

hyper_params <- list( nbins_cats = c(2,4,8, 16) ,
                      learn_rate = seq(0.1,1,0.1),
                      col_sample_rate_per_tree = c(.2, .6, .8, 1)
                       )

grid <- h2o.grid(x = predictors, y = response, training_frame = train, validation_frame = valid,
                 algorithm = "gbm",distribution = "bernoulli", grid_id = "air_grid", hyper_params = hyper_params,
                 stopping_rounds = 5, stopping_tolerance = 1e-4, stopping_metric = "AUC",

                 search_criteria = list(strategy = "Cartesian"), seed = 1234,balance_classes = TRUE,class_sampling_factors = c(0.2,1.0))


###############
#             #
# Exercise 4  #
#             #
###############
h2o.getGrid("air_grid", sort_by="auc", decreasing = TRUE)
best.m <- h2o.getModel("air_grid_model_1")
h2o.auc(h2o.performance(best.m,test))

###############
#             #
# Exercise 5  #
#             #
###############


rfHex <- h2o.randomForest(x=predictors,
                          y= response,
                          ntrees = 100,
                          max_depth = 30,
                          nbins_cats = 16,
                          training_frame= train)


h2o.auc(h2o.performance(rfHex,newdata = valid))
h2o.auc(h2o.performance(rfHex,newdata = test))



###############
#             #
# Exercise 6  #
#             #
###############

ntrees_opts <- c(10000) ## early stopping will stop earlier
max_depth_opts <- seq(1,20)
min_rows_opts <- c(1,5,10,20,50,100)
learn_rate_opts <- seq(0.001,0.01,0.001)
sample_rate_opts <- seq(0.3,1,0.05)
col_sample_rate_opts <- seq(0.3,1,0.05)
col_sample_rate_per_tree_opts = seq(0.3,1,0.05)


hyper_params.rf= list( ntrees = ntrees_opts,
                     max_depth = max_depth_opts,
                     min_rows = min_rows_opts,
                     #learn_rate = learn_rate_opts,
                     sample_rate = sample_rate_opts

)


grid.rf<- h2o.grid(x = predictors, y = response, training_frame = train, validation_frame = valid,
                 algorithm = "randomForest",distribution = "bernoulli", grid_id = "rf_grid_1", hyper_params = hyper_params.rf,
                 stopping_rounds = 5, stopping_tolerance = 1e-4, stopping_metric = "AUC",

                 search_criteria = list(strategy = "Cartesian"), seed = 1234,balance_classes = TRUE)

###############
#             #
# Exercise 7  #
#             #
###############

gbm.en  <- h2o.gbm(x = predictors,
                  y = response,
                  training_frame = train,
                  distribution = "bernoulli",
                  ntrees = 10,
                  max_depth = 3,
                  min_rows = 2,
                  learn_rate = 0.2,
                  nfolds = 5,
                  fold_assignment = "Modulo",
                  keep_cross_validation_predictions = TRUE,
                  seed = 1)

rf.en <- h2o.randomForest(x = predictors,
                          y = response,
                          training_frame = train,
                          ntrees = 50,
                          nfolds = 5,
                          fold_assignment = "Modulo",
                          keep_cross_validation_predictions = TRUE,
                          seed = 1)

# Train a stacked ensemble using the GBM and RF above
ensemble.rf.gbm <- h2o.stackedEnsemble(x = predictors,
                                y = response,
                                training_frame = train,
                                model_id = "en.rf.gbm",
                                base_models = list(gbm.en@model_id, rf.en@model_id))




perf <- h2o.performance(ensemble.rf.gbm, newdata = test)


h2o.auc(perf)
###############
#             #
# Exercise 8  #
#             #
###############

prostate <- h2o.importFile("data\pstate.csv")

prostate.km = h2o.kmeans(prostate, k = 10, x = c("AGE","RACE","GLEASON","CAPSULE","DCAPS") )
print(prostate.km)

###############
#             #
# Exercise 9  #
#             #
###############

prostate.km.2 = h2o.kmeans(prostate, k = 10, x = c("AGE","RACE","GLEASON","CAPSULE","DCAPS") ,init = "Furthest")

par(mfrow = c(1,2))
prostate.ctrs = as.data.frame(prostate.km@model$centers)
plot(prostate.ctrs[,1:2])
plot(prostate.ctrs[,3:4])

h2o.predict(prostate.km,newdata = prostate)

###############
#             #
# Exercise 10  #
#             #
###############



pca_data <- h2o.importFile("data\Train_UWu5bXk.csv")

h2o.str(pca_data)



splits.pca <- h2o.splitFrame(pca_data,c(0.8))
train.pca <- splits.pca[[1]]

test.pca <- splits.pca[[2]]
mkj <- h2o.prcomp(train.pca,k=8,use_all_factor_levels = TRUE)
screeplot(mkj)