h2o_final.R

rm(list=ls())

## Leave your path here to feed in setwd
path_amalesh <- paste0("D:/ALL/Study/KUL/Stats/Amalesh/",
                       "3rd sem/Advanced analyics for business")
path_marc <- paste0("d:/MarcandreC/Desktop/ass2_h2o")

setwd(path_marc)

## The following functions loads the
## dependencies automatically
source("./assigment2_functions_macode.r")
source("./assigment2_macode.r")

## TRUE if you want to install the h20 packages etc.
debug_install <- FALSE

if(debug_install) {
## The following two commands remove any previously installed H2O
## packages for R.
    if ("package:h2o" %in% search()) {
        detach("package:h2o", unload=TRUE)
    }
    if ("h2o" %in% rownames(installed.packages())) {
        remove.packages("h2o")
    }

## Next, we download packages that H2O depends on.
    pkgs <- c("RCurl","jsonlite")
    for (pkg in pkgs) {
        if (!(pkg %in% rownames(installed.packages()))) {
            install.packages(pkg)
        }
    }
## Now we download, install and initialize the H2O package for R.
    install.packages("h2o", type="source",
                     repos=paste0("http://h2o-release.s3.amazonaws.com/",
                                  "h2o/rel-yates/3/R"))
}

## Main meat
## orig.dat must be used for training,
## otherwise validation set are infected
## with rows from training set.
library(h2o)
h2o.init()

train.dat <- orig.dat
test.dat <- test.dat
churn <- Y.orig
train.dat1<- cbind(train.dat,churn)
churn<-read_rds('Y_test.rds')
test.dat1<- cbind(test.dat,churn)
training<- as.h2o(train.dat1)
testing<- as.h2o(test.dat1)
rf_h2o <- h2o.randomForest(
  y = 'churn',
  training_frame = training,
  validation_frame = testing,
  ntrees = 1000,
  stopping_metric = "AUC",    
  stopping_rounds = 10,         
  stopping_tolerance = 0.005,
  seed = 123
)
#modelperformance
h2o.auc(rf_h2o, valid = TRUE)
#automl
h2o_ai<- h2o.automl(y = 'churn',
                training_frame = training,
                validation_frame = testing,
                max_models = 50,stopping_metric = "AUC",
                max_runtime_secs = 1200,seed = 120)
#print all models
lb <- h2o_ai@leaderboard
print(lb, n = nrow(lb))
model_ids <- as.data.frame(h2o_ai@leaderboard$model_id)[,1]
#selecting best ensamble model from leadderboard
se <- h2o.getModel(grep("StackedEnsemble_AllModels", model_ids, value = TRUE)[1])
#summary of best ensamble
se
metalearner <- h2o.getModel(se@model$metalearner$name)
h2o.varimp(metalearner)
h2o.varimp_plot(metalearner)
#gbm
gbm <- h2o.getModel(grep("GBM", model_ids, value = TRUE)[1])
#summary of best gbm
gbm
#knowing parameters used
gbm@allparameters
h2o.varimp(gbm)
h2o.varimp_plot(gbm)

# #explanation
# explainer_rf <- lime::lime(
#   as.data.frame(training[,-36]),
#   model = rf_h2o,
#   bin_continuous = F
# )
# explanation_rf <- lime::explain(
#   as.data.frame(testing[, -36]), 
#   explainer = explainer_rf, 
#   n_labels     = 1, 
#   n_features   = 4)
# klm<-plot_features(explanation_rf,ncol = 5,cases = 63)