Add test for Isolation Forest grid search test in unsupervised fashion.

In my opinion grid search for Isolation Forest should be unsupervised by default and supervised on option.

h2o-r/h2o-package/R/grid.R

@@ -74,11 +74,11 @@ h2o.grid <- function(algorithm,
                      parallelism = 1)
 {
   #Unsupervised algos to account for in grid (these algos do not need response)
-  unsupervised_algos <- c("kmeans", "pca", "svd", "glrm", "extendedisolationforest")
+  unsupervised_algos <- c("kmeans", "pca", "svd", "glrm", "isolationforest", "extendedisolationforest")
   # Parameter list
   dots <- list(...)
   # Add x, y, and training_frame
-  if(!(algorithm %in% c(unsupervised_algos, toupper(unsupervised_algos)))) {
+  if(!(algorithm %in% c(unsupervised_algos, toupper(unsupervised_algos))) || is_supervised) {
     if(!missing(y)) {
       dots$y <- y
     } else {
@@ -94,7 +94,7 @@ h2o.grid <- function(algorithm,
     stop("Must specify training frame, training_frame")
   }
   # If x is missing, then assume user wants to use all columns as features for supervised models only
-  if(!(algorithm %in% c(unsupervised_algos, toupper(unsupervised_algos)))) {
+  if(!(algorithm %in% c(unsupervised_algos, toupper(unsupervised_algos))) || is_supervised) {
     if (missing(x)) {
       if (is.numeric(y)) {
         dots$x <- setdiff(col(training_frame), y)

h2o-r/h2o-package/R/models.R

@@ -4048,6 +4048,7 @@ h2o.sdev <- function(object) {
   if (algo == "kmeans" ||
       algo == "glrm" ||
       algo == "pca" ||
+      algo == "isolationforest" ||
       algo == "extendedisolationforest" ||
       (algo == "deeplearning" && !is.null(params$autoencoder) && params$autoencoder)) {
     FALSE

h2o-r/tests/testdir_algos/isofor/runit_isofor_grid_supervised.R

@@ -5,7 +5,7 @@ source("../../../scripts/h2o-r-test-setup.R")
 
 
 
-test.grid.resume <- function() {
+test.grid.isolationforest.supervised <- function() {
   iris.hex <-
     h2o.importFile(path = locate("smalldata/iris/iris.csv"),
                    destination_frame = "iris.hex")
@@ -21,6 +21,7 @@ test.grid.resume <- function() {
       x = 1:4,
       y = 5,
       training_frame = iris.hex,
+      is_supervised = TRUE,
       hyper_params = hyper_parameters,
       parallelism = 0
     )
@@ -54,9 +55,10 @@ test.grid.resume <- function() {
       training_frame = train,
       validation_frame = test,
       hyper_params = hyper_parameters,
+      is_supervised = TRUE,
       validation_response_column = "label",
       parallelism = 0
     )
 }
 
-doTest("Parallel Grid Search test", test.grid.resume)
+doTest("Parallel Grid Search test", test.grid.isolationforest.supervised)

h2o-r/tests/testdir_algos/isofor/runit_isofor_grid_unsupervised_default.R

@@ -0,0 +1,31 @@
+setwd(normalizePath(dirname(
+  R.utils::commandArgs(asValues = TRUE)$"f"
+)))
+source("../../../scripts/h2o-r-test-setup.R")
+
+
+test.grid.isolationforest.unsupervised.default <- function() {
+  single_blob.hex <-
+    h2o.importFile(path = locate("smalldata/anomaly/single_blob.csv"),
+                   destination_frame = "single_blob.hex")
+
+  ntrees <- c(25, 50, 100)
+  sample_size <- c(64, 128, 256)
+  size_of_hyper_space <- length(ntrees)*length(sample_size)
+
+  hyper_parameters <- list(ntrees = ntrees, sample_size = sample_size)
+  baseline_grid <-
+    h2o.grid(
+      "isolationforest",
+      grid_id = "isofor_grid_test",
+      x = c(1, 2),
+      training_frame = single_blob.hex,
+      hyper_params = hyper_parameters,
+      parallelism = 0
+    )
+  print(paste("Expected size of hyperparameter space is", length(baseline_grid@model_ids)))
+  expect_equal(length(baseline_grid@model_ids), size_of_hyper_space)
+}
+
+
+doTest("Parallel Grid Search test for Isolation Forest - default grid search is in unsupervised fashion", test.grid.isolationforest.unsupervised.default)