Fix categorical encoding

Author: maurever

h2o-algos/src/main/java/hex/knn/KNN.java

@@ -41,6 +41,10 @@ public boolean isSupervised() {
         if( null == _parms._distance) {
             error("_distance", "Distance parameter not set.");
         }
+        if (null != _parms._categorical_encoding && Model.Parameters.CategoricalEncodingScheme.Enum != _parms._categorical_encoding 
+                && Model.Parameters.CategoricalEncodingScheme.AUTO != _parms._categorical_encoding) {
+            error("_categorical_encoding", "Only enum categorical encoding is supported.");
+        }
     }
 
     class KNNDriver extends Driver {

h2o-algos/src/main/java/hex/knn/KNNModel.java

@@ -76,10 +76,12 @@ public ModelMetrics.MetricBuilder makeMetricBuilder(String[] domain) {
 
     @Override
     protected double[] score0(double[] data, double[] preds) {
-        Frame train = _parms._train.get(); 
+        Frame train = new Frame(_parms.train());
+        adaptTestForTrain(train, false, false);
         int idIndex = train.find(_parms._id_column);
         int responseIndex = train.find(_parms._response_column);
         byte idType = train.types()[idIndex];
+        
         preds = new KNNScoringTask(data, _parms._k, _output.nclasses(), KNNDistanceFactory.createDistance(_parms._distance), idIndex, idType, 
                                     responseIndex).doAll(train).score();
         Scope.untrack(train);

h2o-py/tests/testdir_algos/knn/pyunit_knn_categorical_data.py

@@ -0,0 +1,69 @@
+import sys, os
+
+sys.path.insert(1, os.path.join("..", "..", ".."))
+import h2o
+from tests import pyunit_utils, assert_equals
+from h2o.estimators.knn import H2OKnnEstimator
+import numpy as np
+
+
+def knn_categorical_data():
+    seed = 12345
+    id_column = "id"
+    response_column = "IsDepDelayed"
+    x_names = ["Origin", "Distance", "Dest"]
+
+    train = h2o.upload_file(pyunit_utils.locate("smalldata/testng/airlines_train.csv"))
+    train[response_column] = train[response_column].asfactor()
+    train[id_column] = h2o.H2OFrame(np.arange(0, train.shape[0]))
+    print(train.shape)
+    train = train[0:1000, :]
+
+    # test AUTO categorical encoding 
+    model = H2OKnnEstimator(
+        k=3,
+        id_column=id_column,
+        distance="euclidean",
+        seed=seed,
+        auc_type="macroovr"
+    )
+    
+    model.train(y=response_column, x=x_names, training_frame=train)
+    assert model is not None
+    
+    preds = model.predict(train)
+    assert preds is not None
+
+    # test enum categorical encoding 
+    model = H2OKnnEstimator(
+        k=3,
+        id_column=id_column,
+        distance="euclidean",
+        seed=seed,
+        auc_type="macroovr",
+        categorical_encoding="enum"
+    )
+
+    model.train(y=response_column, x=x_names, training_frame=train)
+
+    # test different categorical encoding than enum
+    try:
+        model = H2OKnnEstimator(
+            k=3,
+            id_column=id_column,
+            distance="euclidean",
+            seed=seed,
+            auc_type="macroovr",
+            categorical_encoding="one_hot_explicit")
+
+        model.train(y=response_column, x=x_names, training_frame=train)
+    except Exception as ex:
+        exception = str(ex)
+        assert ("H2OModelBuilderIllegalArgumentException" in exception)
+        assert ("_categorical_encoding: Only enum categorical encoding is supported." in exception)
+        
+
+if __name__ == "__main__":
+    pyunit_utils.standalone_test(knn_categorical_data)
+else:
+    knn_categorical_data()

h2o-py/tests/testdir_algos/knn/pyunit_knn_smoke.py

@@ -13,9 +13,9 @@ def knn_api_smoke():
     response_column = "class"
     x_names = ["sepal_len", "sepal_wid", "petal_len", "petal_wid"]
 
-    train_h2o = h2o.upload_file(pyunit_utils.locate("smalldata/iris/iris_wheader.csv"))
-    train_h2o[response_column] = train_h2o[response_column].asfactor()
-    train_h2o[id_column] = h2o.H2OFrame(np.arange(0, train_h2o.shape[0]))
+    train = h2o.upload_file(pyunit_utils.locate("smalldata/iris/iris_wheader.csv"))
+    train[response_column] = train[response_column].asfactor()
+    train[id_column] = h2o.H2OFrame(np.arange(0, train.shape[0]))
 
     model = H2OKnnEstimator(
         k=3,
@@ -24,7 +24,7 @@ def knn_api_smoke():
         seed=seed,
         auc_type="macroovr"
     )
-    model.train(y=response_column, x=x_names, training_frame=train_h2o)
+    model.train(y=response_column, x=x_names, training_frame=train)
     perf = model.model_performance()
 
     print(perf)
@@ -37,6 +37,13 @@ def knn_api_smoke():
     distances = model.distances()
     assert distances is not None
 
+    preds = model.predict(train)
+    assert preds is not None
+    
+    print(preds[0,:])
+    print(preds[51, :])
+    print(preds[101, :])
+
 
 if __name__ == "__main__":
     pyunit_utils.standalone_test(knn_api_smoke)