Resolve merge conflict [nocheck]

h2o-algos/src/main/java/hex/tree/gbm/GBMModel.java

@@ -12,6 +12,7 @@
 import water.fvec.NewChunk;
 import water.fvec.Vec;
 import water.util.FrameUtils;
+import water.util.Log;
 import water.util.TwoDimTable;
 
 import java.util.*;
@@ -287,6 +288,7 @@ public FeatureInteractions getFeatureInteractions(int maxInteractionDepth, int m
       for (int j = 0; j < nclasses; j++) {
         FeatureInteractions currentTreeFeatureInteractions = new FeatureInteractions();
         SharedTreeSubgraph tree = this.getSharedTreeSubgraph(i, j);
+
         List<SharedTreeNode> interactionPath = new ArrayList<>();
         Set<String> memo = new HashSet<>();
 
@@ -295,7 +297,10 @@ public FeatureInteractions getFeatureInteractions(int maxInteractionDepth, int m
         featureInteractions.mergeWith(currentTreeFeatureInteractions);
       }
     }
-
+    if(featureInteractions.isEmpty()){
+      Log.warn("There is no feature interaction for this model.");
+      return null;
+    }
     return featureInteractions;
   }
 

h2o-core/src/main/java/hex/FeatureInteractions.java

@@ -41,12 +41,18 @@ public void mergeWith(FeatureInteractions featureInteractions) {
             }
         }
     }
+
+    public boolean isEmpty(){
+        return entrySet().isEmpty();
+    }
 
     public int maxDepth() {
+        if(isEmpty()) return 0;
         return Collections.max(this.entrySet(), Comparator.comparingInt(entry -> entry.getValue().depth)).getValue().depth;
     }
 
     public TwoDimTable[] getAsTable() {
+        if(isEmpty()) return  null;
         int maxDepth = maxDepth();
         TwoDimTable[] twoDimTables = new TwoDimTable[maxDepth + 1];
         for (int depth = 0; depth < maxDepth + 1; depth++) {
@@ -55,23 +61,24 @@ public TwoDimTable[] getAsTable() {
         return twoDimTables;
     }
 
-    List<FeatureInteraction> getFeatureInteractionsOfDepth(int depthRequired) {
+    private List<FeatureInteraction> getFeatureInteractionsOfDepth(int depthRequired) {
         return this.entrySet()
                 .stream()
                 .filter(entry -> entry.getValue().depth == depthRequired)
                 .map(Map.Entry::getValue)
                 .collect(Collectors.toList());
     }
 
-    List<FeatureInteraction> getFeatureInteractionsWithLeafStatistics() {
+    private List<FeatureInteraction> getFeatureInteractionsWithLeafStatistics() {
         return this.entrySet()
                 .stream()
                 .filter(entry -> entry.getValue().hasLeafStatistics == true)
                 .map(Map.Entry::getValue)
                 .collect(Collectors.toList());
     }
 
-    TwoDimTable constructFeatureInteractionsTable(int depth) {
+    private TwoDimTable constructFeatureInteractionsTable(int depth) {
+        assert depth >= 0  : "Depth has to be >= 0.";
         String[] colHeaders = new String[] {"Interaction", "Gain", "FScore", "wFScore", "Average wFScore", "Average Gain", 
                 "Expected Gain", "Gain Rank", "FScore Rank", "wFScore Rank", "Avg wFScore Rank", "Avg Gain Rank", 
                 "Expected Gain Rank", "Average Rank", "Average Tree Index", "Average Tree Depth"};
@@ -134,7 +141,7 @@ TwoDimTable constructFeatureInteractionsTable(int depth) {
         return table;
     }
 
-    int indexOfInteractionWithName(String name, List<FeatureInteraction> featureInteractions) {
+    private int indexOfInteractionWithName(String name, List<FeatureInteraction> featureInteractions) {
         for (int i = 0; i < featureInteractions.size(); i++)
             if (featureInteractions.get(i).name == name)
                 return i;
@@ -182,7 +189,7 @@ public TwoDimTable[] getSplitValueHistograms() {
         return splitValueHistograms;
     }
 
-    TwoDimTable constructHistogramForFeatureInteraction(FeatureInteraction featureInteraction) {
+    private TwoDimTable constructHistogramForFeatureInteraction(FeatureInteraction featureInteraction) {
         String[] colHeaders = new String[] {"Split Value", "Count"};
         String[] colTypes = new String[] {"double", "int"};
         String[] colFormat = new String[] {"%.5f", "%d"};
@@ -297,11 +304,13 @@ public static void collectFeatureInteractions(SharedTreeNode node, List<SharedTr
     }
 
     public static TwoDimTable[][] getFeatureInteractionsTable(FeatureInteractions featureInteractions) {
+        if(featureInteractions == null) {
+            return null;
+        }
         TwoDimTable[][] table = new TwoDimTable[3][];
-        table[0] =  featureInteractions.getAsTable();
+        table[0] = featureInteractions.getAsTable();
         table[1] = new TwoDimTable[]{featureInteractions.getLeafStatisticsTable()};
         table[2] = featureInteractions.getSplitValueHistograms();
-
         return table;
     }
 }

h2o-core/src/main/java/water/api/ModelsHandler.java

@@ -179,7 +179,9 @@ public FeatureInteractionV3 makeFeatureInteraction(int version, FeatureInteracti
     Model model = getFromDKV("key", s.model_id.key());
     if (model instanceof FeatureInteractionsCollector) {
       TwoDimTable[][] featureInteractions = ((FeatureInteractionsCollector) model).getFeatureInteractionsTable(s.max_interaction_depth, s.max_tree_depth, s.max_deepening);
-
+      if(featureInteractions == null){
+        return s;
+      }
       s.feature_interaction = new TwoDimTableV3[featureInteractions[0].length + featureInteractions[2].length + 1];
 
       for (int i = 0; i < featureInteractions[0].length; i++) {
@@ -189,7 +191,6 @@ public FeatureInteractionV3 makeFeatureInteraction(int version, FeatureInteracti
       for (int i = 0; i < featureInteractions[2].length; i++) {
         s.feature_interaction[i + featureInteractions[0].length + 1] = new TwoDimTableV3().fillFromImpl(featureInteractions[2][i]);
       }
-
       return s;
     } else {
       throw H2O.unimpl(String.format("%s does not support feature interactions calculation", model._parms.fullName()));

h2o-core/src/test/java/hex/FeatureInteractionsTest.java

@@ -224,5 +224,12 @@ public void testGetFeatureInteractionsTable() {
         expected.add("AA|CC");
         assertTrue(vars.containsAll(expected));
     }
+
+    @Test
+    public void testNoInteractions() {
+        FeatureInteractions featureInteractions = new FeatureInteractions();
+        TwoDimTable[] table = featureInteractions.getAsTable();
+        assert table == null;
+    }
 
 }

h2o-docs/src/product/data-science/algo-params/custom_metric_func.rst

@@ -3,7 +3,7 @@
 ``custom_metric_func``
 ----------------------
 
-- Available in: GBM, DRF, Deeplearning, Stacked Ensembles
+- Available in: GBM, DRF, Deeplearning, Stacked Ensembles, GLM
 - Hyperparameter: no
 
 Description

h2o-docs/src/product/data-science/algo-params/upload_custom_metric.rst

@@ -3,7 +3,7 @@
 ``upload_custom_metric``
 ------------------------
 
-- Available in: GBM, DRF, Deeplearning
+- Available in: GBM, DRF, Deeplearning, GLM
 - Hyperparameter: no
 
 Description

h2o-docs/src/product/data-science/glm.rst

@@ -43,6 +43,8 @@ Algorithm-specific parameters
 
 -  **calc_like**: Specify whether to return likelihood function value for HGLM or normal GLM. Setting this option to ``True`` while disabling ``HGLM`` will enable the calculation of the full log likelihood and full AIC. This option defaults to ``False`` (disabled). 
 
+- `custom_metric_func <algo-params/custom_metric_func.html>`__: Specify a custom evaluation function.
+
 - **dispersion_epsilon**: If changes in dispersion parameter estimation or loglikelihood value is smaller than ``dispersion_epsilon``, this will break out of the dispersion parameter estimation loop using maximum likelihood. This option defaults to ``0.0001``.
 
 - **dispersion_learning_rate**: (Applicable only when ``dispersion_parameter_method="ml"``) This value controls how much the dispersion parameter estimate will be changed when the calculated loglikelihood actually decreases with the new dispersion. In this case, instead of setting *dispersion = dispersion + change*, it is *dispersion + dispersion_learning_rate* :math:`\times` *change*. This option must be > 0 and defaults to ``0.5``.
@@ -65,6 +67,8 @@ Algorithm-specific parameters
 
 -  `rand_family <algo-params/rand_family.html>`__: The Random Component Family specified as an array. You must include one family for each random component. Currently only ``rand_family=["gaussisan"]`` is supported.
 
+- `upload_custom_metric <algo-params/upload_custom_metric.html>`__: Upload a custom metric into a running H2O cluster.
+
 HGLM parameters
 '''''''''''''''
 

h2o-docs/src/product/getting-data-into-h2o.rst

@@ -8,7 +8,7 @@ The first step toward building and scoring your models is getting your data into
 Supported File Formats
 ----------------------
 
-H2O currently supports the following file types:
+H2O supports the following file types:
 
 - CSV (delimited, UTF-8 only) files (including GZipped CSV)
 - ORC
@@ -18,6 +18,7 @@ H2O currently supports the following file types:
 - XLSX (BIFF 8 only)
 - Avro version 1.8.0 (without multifile parsing or column type modification)
 - Parquet
+- Google Storage (gs://)
 
 **Notes**: 
 

h2o-py/docs/data.rst

@@ -40,6 +40,7 @@ The following data sources are supported:
 * A Directory (with many data files inside at the *same* level -- no support for recursive import of data)
 * S3/S3N
 * Native Language Data Structure (c.f. the subsequent section)
+* Google Storage (gs://)
 
 .. code-block:: python
 

h2o-py/h2o/h2o.py

@@ -413,7 +413,7 @@ def import_file(path=None, destination_frame=None, parse=True, header=0, sep=Non
                 na_strings=None, pattern=None, skipped_columns=None, custom_non_data_line_markers=None,
                 partition_by=None, quotechar=None, escapechar=None):
     """
-    Import a dataset that is already on the cluster.
+    Import files into an H2O cluster. The default behavior is to pass-through to the parse phase automatically.
 
     The path to the data must be a valid path for each node in the H2O cluster. If some node in the H2O cluster
     cannot see the file, then an exception will be thrown by the H2O cluster. Does a parallel/distributed

h2o-py/h2o/model/model_base.py

@@ -629,10 +629,13 @@ def feature_interaction(self, max_interaction_depth=100, max_tree_depth=100, max
         # For now, redirect to h2o.model.extensions.feature_interaction for models that support the feature, and print legacy message for others..
         # Later, the method will be exposed only for models supporting the feature.
         if has_extension(self, 'FeatureInteraction'):
-            return self._feature_interaction(max_interaction_depth=max_interaction_depth, 
-                                             max_tree_depth=max_tree_depth, 
-                                             max_deepening=max_deepening, 
-                                             path=path)
+            table = self._feature_interaction(max_interaction_depth=max_interaction_depth,
+                                              max_tree_depth=max_tree_depth,
+                                              max_deepening=max_deepening,
+                                              path=path)
+            if table is None:
+                print("There is no feature interaction for this model.")
+            return table
         print("No calculation available for this model")
 
     def h(self, frame, variables):

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

@@ -3225,7 +3225,10 @@ h2o.feature_interaction <- function(model, max_interaction_depth = 100, max_tree
 
             json <- .h2o.doSafePOST(urlSuffix = "FeatureInteraction", parms=parms)
             source <- .h2o.fromJSON(jsonlite::fromJSON(json,simplifyDataFrame=FALSE))
-
+            if(is.null(source$feature_interaction)){
+                warning(paste0("There is no feature interaction for this model."))
+                return(NULL)
+            }
             return(source$feature_interaction)
         } else {
             warning(paste0("No calculation available for this model"))

h2o-r/tests/testdir_algos/gbm/runit_GBM_feature_interaction_cv.R

@@ -0,0 +1,29 @@
+setwd(normalizePath(dirname(R.utils::commandArgs(asValues=TRUE)$"f")))
+source("../../../scripts/h2o-r-test-setup.R")
+
+library(ggplot2)
+
+test.feature_interaction_with_cv <- function() {
+    diamonds <- ggplot2::diamonds
+    diamonds$cut <- factor(diamonds$cut, ordered = FALSE)
+    diamonds$color <- factor(diamonds$color, ordered = FALSE)
+    diamonds$clarity <- factor(diamonds$clarity, ordered = FALSE)
+    diamonds <- as.h2o(diamonds)
+    diamonds$expensive <- h2o.asfactor(ifelse(diamonds$price == 5000, 1, 0))
+
+    train <- diamonds
+    train$fold <- h2o.kfold_column(data = train, nfolds = 3, seed = 123)
+
+    params <- list( x = setdiff(names(diamonds), "expensive"), y = "expensive", fold_column = "fold", training_frame = as.name("train"), validation_frame = NULL, distribution = "bernoulli", learn_rate = 0.1, ntrees = 500, min_split_improvement = 1e-3, stopping_rounds = 3, stopping_tolerance = 0.001, seed = 456 )
+    my_gbm <- do.call(what = "h2o.gbm", args = params)
+
+    # feature interaction with main model
+    print(h2o.feature_interaction(model = my_gbm))
+
+    # feature interaction with cv model where tree depth = 0
+    my_cv_gbm <- h2o.getModel(my_gbm@model$cross_validation_models[[1]]$name)
+    fi <-h2o.feature_interaction(model = my_cv_gbm)
+    expect_true(is.null(fi))
+}
+
+doTest("Test feature interaction with CV enabled", test.feature_interaction_with_cv)