Add R script for Next.ML.

R/examples/H2ODeepLearningNextML.R

@@ -0,0 +1,104 @@
+## Install H2O R package (same version as H2O Server)
+if ("package:h2o" %in% search()) { detach("package:h2o", unload=TRUE) }
+if ("h2o" %in% rownames(installed.packages())) { remove.packages("h2o") }
+install.packages("h2o", repos=(c("http://h2o-release.s3.amazonaws.com/h2o/rel-mirzakhani/2/R", getOption("repos"))))
+
+## Load H2O R libary
+library(h2o)
+
+## Connect to H2O Server
+h2oServer <- h2o.init(ip="localhost", port=8996)
+
+
+
+## HIGGS
+
+## Parse Higgs dataset
+higgs_hex <- h2o.importFile(h2oServer, "higgs.100k.csv.gz")
+summary(higgs_hex)
+dim(higgs_hex)
+
+## Split data into train/validation and store under user-given keys in the H2O Store
+random <- h2o.runif(higgs_hex, seed = 123456789)
+higgs_train_hex <- h2o.assign(higgs_hex[random < .75,], "higgs_train_hex")
+higgs_validation_hex  <- h2o.assign(higgs_hex[random >= .75,], "higgs_validation_hex")
+
+## Clean up temporaries
+h2o.rm(h2oServer, grep(pattern = "Last.value", x = h2o.ls(h2oServer)$Key, value = TRUE))
+
+
+## Run DL model
+response = 1
+low_level_predictors = c(2:22)
+high_level_predictors = c(23:29)
+
+help(h2o.deeplearning)
+
+ll_model <- h2o.deeplearning(data=higgs_train_hex, validation=higgs_validation_hex, 
+                             x=low_level_predictors, y=response)
+ll_model
+
+hl_model <- h2o.deeplearning(data=higgs_train_hex, validation=higgs_validation_hex, 
+                             x=high_level_predictors, y=response)
+hl_model
+
+ll_hl_model <- h2o.deeplearning(data=higgs_train_hex, validation=higgs_validation_hex, 
+                                x=c(low_level_predictors,high_level_predictors), y=response)
+ll_hl_model
+
+## Example how to make and export predictions
+predictions <- h2o.predict(hl_model, higgs_validation_hex)
+head(predictions)
+h2o.exportFile(data=predictions, path="predictions")
+
+
+
+## ADULT
+
+## Run grid search on Adult
+adult_hex <- h2o.importFile(h2oServer, "adult.gz")
+dim(adult_hex)
+summary(adult_hex)
+
+## Set column names manually
+colnames(adult_hex) <- c("age","workclass","fnlwgt","education","education-num","marital-status",
+                         "occupation","relationship","race","sex","capital-gain","capital-loss",
+                         "hours-per-week","native-country","income")
+summary(adult_hex)
+
+## Turn age into a categorical factor
+adult_hex$age <- as.factor(adult_hex$age)
+summary(adult_hex)
+
+pred = 1:14
+adult_grid <- h2o.deeplearning(data=adult_hex, x=pred, y="income",
+                               activation=c("Rectifier"), l1=c(0,1e-6), l2=0,
+                               hidden=list(c(20,20,20), c(50,50)), epochs=10)
+
+## Find the best model (based on training error, since no validation dataset specified)
+best_model <- adult_grid@model[[1]]
+best_model
+best_model@model$params$activation
+best_model@model$params$l1
+best_model@model$params$hidden
+
+
+## Compare to GBM
+gbm_model <- h2o.gbm(data=adult_hex, x=pred, y=resp)
+gbm_model
+
+
+
+## MNIST
+## World-record parameters at http://learn.h2o.ai/content/hands-on_training/deep_learning.html
+
+mnist_train_hex <- h2o.importFile(h2oServer, "mnist.train.csv.gz")
+mnist_test_hex <- h2o.importFile(h2oServer, "mnist.test.csv.gz")
+dim(mnist_train_hex)
+dim(mnist_test_hex)
+
+mnist_model <- h2o.deeplearning(data=mnist_train_hex, validation=mnist_test_hex, hidden=c(20,20,20), epochs=1, x=1:784, y=785)
+mnist_model
+
+
+## For more examples, see http://learn.h2o.ai/content/