wisc_diag_breast_cancer_h2o_demo.R

library(h2o)
localH2O <- h2o.init(nthreads = -1)  #Start up H2O cluster using nthreads = ncores


# Get training data:
data <- h2o.importFile("http://www.stat.berkeley.edu/~ledell/data/wisc-diag-breast-cancer-shuffled.csv",
  destination_frame = "breast_cancer")
y <- "diagnosis"  #Response column
x <- setdiff(names(data), c(y, "id"))  #remove 'id' and response col


# Train & Test
set.seed(1)
ss <- h2o.splitFrame(data)  #split data into train & test partitions
training_frame <- ss[[1]]
validation_frame <- ss[[2]]

# Train a GLM
h2o_glm <- h2o.glm(x = x, y = y, 
                   training_frame = training_frame, 
                   validation_frame = validation_frame,
                   family = "binomial")
print(h2o_glm@model$validation_metrics@metrics$AUC)
# Test set AUC: 0.9935432

h2o.auc(h2o_glm, valid = TRUE)  #utility function to get AUC



# Cross-validated GBM
h2o_gbm <- h2o.gbm(x = x, y = y, 
                   training_frame = data, 
                   nfolds = 5,
                   family = "binomial", 
                   seed = 1)
print(h2o_gbm@model$cross_validation_metrics@metrics$AUC)
# CV AUC:  0.9894099


# Cross-validate a Random Forest
h2o_rf <- h2o.randomForest(x = x, y = y, 
                   training_frame = data, 
                   nfolds = 5,
                   family = "binomial",
                   seed = 1)
print(h2o_rf@model$cross_validation_metrics@metrics$AUC)
# CV AUC:  0.9902621