gistfile1.txt

#Random hyper parameter search
models <- c()
for (i in 1:20) {
  rand_activation <- c("TanhWithDropout", "RectifierWithDropout")[sample(1:2,1)]
  rand_numlayers <- sample(2:5,1)
  rand_hidden <- c(sample(10:50,rand_numlayers,T))
  rand_l1 <- runif(1, 0, 1e-3)
  rand_l2 <- runif(1, 0, 1e-3)
  rand_dropout <- c(runif(rand_numlayers, 0, 0.6))
  rand_input_dropout <- runif(1, 0, 0.5)
  dlmodel <- h2o.deeplearning(
    x = 1:369, y = 371, 
    training_frame = train_holdout.hex,
    validation_frame = valid_holdout.hex, 
    epochs=1,
    stopping_metric="misclassification",
    stopping_tolerance=1e-2,        ## stop when logloss does not improve by >=1% for 2 scoring events
    stopping_rounds=2,
    score_validation_samples=10000, ## downsample validation set for faster scoring
    score_duty_cycle=0.025,         ## don't score more than 2.5% of the wall time
    max_w2=10,                      ## can help improve stability for Rectifier
    ### Random parameters
    activation=rand_activation, 
    hidden=rand_hidden, 
    l1=rand_l1, l2=rand_l2,
    input_dropout_ratio=rand_input_dropout, 
    hidden_dropout_ratios=rand_dropout
  )                                
  models <- c(models, dlmodel)
}

# Find the best model (lowest mse on the validation holdout set)
base_auc <- 0.5
for (i in 1:length(models)) {
  auc <- h2o.auc( h2o.performance(models[[i]], valid_holdout.hex))
  if (auc > base_auc) {
    base_auc <- auc
    best_model <- models[[i]]
  }
}