PUBDEV-8547 - eif scoring history - test score_each_iteration and sco…

…re_tree_interval in java
h2oai · Oct 26, 2023 · e2f32ac · e2f32ac
1 parent 669aca2
commit e2f32ac
Showing 1 changed file with 110 additions and 0 deletions.
diff --git a/h2o-algos/src/test/java/hex/tree/isoforextended/ExtendedIsolationForestTest.java b/h2o-algos/src/test/java/hex/tree/isoforextended/ExtendedIsolationForestTest.java
@@ -382,4 +382,114 @@ public void reduce(MeanScoreTask other) {
             totalMeanLength += other.totalMeanLength;
         }
     }
+
+    @Test
+    public void testScoreEachIteration() {
+        try {
+            Scope.enter();
+            Frame train = Scope.track(parseTestFile("smalldata/anomaly/single_blob.csv"));
+            ExtendedIsolationForestModel.ExtendedIsolationForestParameters p =
+                    new ExtendedIsolationForestModel.ExtendedIsolationForestParameters();
+            p._train = train._key;
+            p._seed = 0xDECAF;
+            p._ntrees = 5;
+            p._score_each_iteration = true;
+            p._extension_level = train.numCols() - 1;
+
+            ExtendedIsolationForest eif = new ExtendedIsolationForest(p);
+            ExtendedIsolationForestModel model = eif.trainModel().get();
+            assertNotNull(model);
+            Scope.track_generic(model);
+
+            LOG.info(model._output._scoring_history);
+
+            assertEquals("Number of rows is not correct", 6, model._output._scoring_history.getRowDim());
+            for (int treeNum = 0; treeNum <= p._ntrees; treeNum++) {
+                assertEquals("Tree number is not correct", treeNum, model._output._scoring_history.get(treeNum, 2));
+            }
+        } finally {
+            Scope.exit();
+        }
+    }
+
+    @Test
+    public void testScoreTreeIntervalSmoke() {
+        try {
+            Scope.enter();
+            Frame train = Scope.track(parseTestFile("smalldata/anomaly/single_blob.csv"));
+            ExtendedIsolationForestModel.ExtendedIsolationForestParameters p =
+                    new ExtendedIsolationForestModel.ExtendedIsolationForestParameters();
+            p._train = train._key;
+            p._seed = 0xDECAF;
+            p._ntrees = 5;
+            p._score_tree_interval = 2;
+            p._extension_level = train.numCols() - 1;
+
+            ExtendedIsolationForest eif = new ExtendedIsolationForest(p);
+            ExtendedIsolationForestModel model = eif.trainModel().get();
+            assertNotNull(model);
+            Scope.track_generic(model);
+
+            LOG.info(model._output._scoring_history);
+
+            assertEquals("Number of rows is not correct", 4, model._output._scoring_history.getRowDim());
+            assertEquals("Tree number is not correct", 0, model._output._scoring_history.get(0, 2));
+            assertEquals("Tree number is not correct", 2, model._output._scoring_history.get(1, 2));
+            assertEquals("Tree number is not correct", 4, model._output._scoring_history.get(2, 2));
+            assertEquals("Tree number is not correct", 5, model._output._scoring_history.get(3, 2));
+        } finally {
+            Scope.exit();
+        }
+    }
+
+    @Test
+    public void testScoreTreeInterval() {
+        try {
+            Scope.enter();
+            Frame train = Scope.track(parseTestFile("smalldata/anomaly/single_blob.csv"));
+            ExtendedIsolationForestModel.ExtendedIsolationForestParameters p =
+                    new ExtendedIsolationForestModel.ExtendedIsolationForestParameters();
+            p._train = train._key;
+            p._seed = 0xDECAF;
+            p._ntrees = 100;
+            p._score_tree_interval = 30;
+            p._extension_level = train.numCols() - 1;
+
+            ExtendedIsolationForest eif = new ExtendedIsolationForest(p);
+            ExtendedIsolationForestModel model = eif.trainModel().get();
+            assertNotNull(model);
+            Scope.track_generic(model);
+
+            LOG.info(model._output._scoring_history);
+
+            p._ntrees = 30;
+            p._score_tree_interval = 0;
+            eif = new ExtendedIsolationForest(p);
+            ExtendedIsolationForestModel model30 = eif.trainModel().get();
+            assertNotNull(model30);
+            Scope.track_generic(model30);
+            ModelMetricsAnomaly modelMetricsAnomaly30 = (ModelMetricsAnomaly) model30._output._training_metrics;
+            p._ntrees = 60;
+            eif = new ExtendedIsolationForest(p);
+            ExtendedIsolationForestModel model60 = eif.trainModel().get();
+            assertNotNull(model60);
+            Scope.track_generic(model60);
+            ModelMetricsAnomaly modelMetricsAnomaly60 = (ModelMetricsAnomaly) model60._output._training_metrics;
+            p._ntrees = 90;
+            eif = new ExtendedIsolationForest(p);
+            ExtendedIsolationForestModel model90 = eif.trainModel().get();
+            assertNotNull(model90);
+            Scope.track_generic(model90);
+            ModelMetricsAnomaly modelMetricsAnomaly90 = (ModelMetricsAnomaly) model90._output._training_metrics;
+
+            assertEquals("Partial score is not correct", modelMetricsAnomaly30._mean_score, model._output._scoring_history.get(1, 3));
+            assertEquals("Partial score is not correct", modelMetricsAnomaly30._mean_normalized_score, model._output._scoring_history.get(1, 4));
+            assertEquals("Partial score is not correct", modelMetricsAnomaly60._mean_score, model._output._scoring_history.get(2, 3));
+            assertEquals("Partial score is not correct", modelMetricsAnomaly60._mean_normalized_score, model._output._scoring_history.get(2, 4));
+            assertEquals("Partial score is not correct", modelMetricsAnomaly90._mean_score, model._output._scoring_history.get(3, 3));
+            assertEquals("Partial score is not correct", modelMetricsAnomaly90._mean_normalized_score, model._output._scoring_history.get(3, 4));
+        } finally {
+            Scope.exit();
+        }
+    }
 }