deep_learning

library(tidyverse)
library(mlbench)
library(recipes)
library(h2o)

data("Ionosphere")
str(Ionosphere)

##Build a Deep learning Algorithim to Ionosphere

rec <- recipe(Class ~ . , data = Ionosphere) %>% 
  step_YeoJohnson(all_numeric()) %>% 
  step_center(all_numeric()) %>% 
  step_scale(all_numeric()) %>% 
  step_corr(all_numeric(), threshold = 0.9) %>% 
  step_nzv(all_predictors()) %>% 
  step_dummy(all_nominal(), -all_outcomes())


preparo <- prep(rec, Ionosphere)
Ionosphere.h2o <- bake(preparo, Ionosphere)
str(Ionosphere.h2o)
table(Ionosphere.h2o$Class)

####h2o init ########################################################
h2o.init(nthreads = -1, max_mem_size = '6g')


##Clean h2o
h2o.removeAll()

##Load Data
iono <- as.h2o(Ionosphere.h2o, destination_frame = 'Ionosphere')

##Split
split <- h2o.splitFrame(iono, ratios = 0.8)
iono.train <- split[[1]]
iono.test <- split[[2]]

#Set variables
x <- setdiff(names(iono.train), 'Class')
y <- 'Class'

iono.deep <- h2o.deeplearning(x = x, y = y, training_frame = iono.train,
                              nfolds = 5, model_id = 'iono01',
                              hidden = c(200,200),
                              activation = 'RectifierWithDropout',
                              hidden_dropout_ratios = c(0.5, 0.3),
                              input_dropout_ratio = 0.3,
                              l1 = 0.0005,
                              l2 = 0.0005,
                              stopping_metric = 'AUC',
                              stopping_tolerance = 0.001,
                              stopping_rounds = 4,
                              epochs = 1000)

h2o.performance(iono.deep, newdata = iono.test)
plot(iono.deep)

##Achive an acuracy of 0.9967 in test !!!