[GH-5720] AIC/Loglikelihood test fixes (#5723)

* fix metrics generation (#5718) (cherry picked from commit 3af804b) * [GH-5720] Add AIC/Loglikelihood metrics to MetricsTestSuite (#5721) * fix metrics generation (#5718) (cherry picked from commit 3af804b) * [GH-5720] Add AIC/Loglikelihood metrics to MetricsTestSuite (cherry picked from commit b3bef8b) * [GH-5720] fix H2oGridSearchTestSuite NaN != NaN * [GH-5720] fix H2oGridSearchTestSuite NaN != NaN * aic/loglikelihood optional to be backwards compatible * fix metrics in test_gridsearch.py * fix test_mojo.py nan == nan
h2oai · Apr 8, 2024 · edbd884 · edbd884
1 parent b3bef8b
commit edbd884
Show file tree

Hide file tree

Showing 4 changed files with 12 additions and 7 deletions.
diff --git a/...eration/src/main/scala/ai/h2o/sparkling/api/generation/common/MetricFieldExceptions.scala b/...eration/src/main/scala/ai/h2o/sparkling/api/generation/common/MetricFieldExceptions.scala
@@ -21,5 +21,6 @@ object MetricFieldExceptions {
   def ignored(): Set[String] =
     Set("__meta", "domain", "model", "model_checksum", "frame", "frame_checksum", "model_category", "predictions")
 
-  def optional(): Set[String] = Set("custom_metric_name", "custom_metric_value", "mean_score", "mean_normalized_score")
+  def optional(): Set[String] =
+    Set("custom_metric_name", "custom_metric_value", "mean_score", "mean_normalized_score", "AIC", "loglikelihood")
 }
diff --git a/ml/src/test/scala/ai/h2o/sparkling/ml/algos/H2OGridSearchTestSuite.scala b/ml/src/test/scala/ai/h2o/sparkling/ml/algos/H2OGridSearchTestSuite.scala
@@ -23,10 +23,9 @@ import ai.h2o.sparkling.{SharedH2OTestContext, TestUtils}
 import hex.Model
 import hex.tree.gbm.GBMModel.GBMParameters
 import org.apache.spark.sql.functions._
-import org.apache.spark.sql.DataFrame
+import org.apache.spark.sql.{DataFrame, Row, SparkSession}
 import org.apache.spark.ml.param.{ParamMap, Params}
 import org.apache.spark.ml.{Pipeline, PipelineModel}
-import org.apache.spark.sql.SparkSession
 import org.junit.runner.RunWith
 import org.scalatest.junit.JUnitRunner
 import org.scalatest.{FunSuite, Matchers}
@@ -268,6 +267,8 @@ class H2OGridSearchTestSuite extends FunSuite with Matchers with SharedH2OTestCo
   }
 
   test("The first row returned by getGridModelsMetrics() method is the same as current metrics of the best model") {
+    import spark.implicits._
+
     val drf = new H2ODRF()
       .setFeaturesCols(Array("AGE", "RACE", "DPROS", "DCAPS", "PSA", "VOL", "GLEASON"))
       .setLabelCol("CAPSULE")
@@ -288,7 +289,8 @@ class H2OGridSearchTestSuite extends FunSuite with Matchers with SharedH2OTestCo
     val gridModelMetrics = search.getGridModelsMetrics().drop("MOJO Model ID")
     val modelMetricsFromGrid = gridModelMetrics.columns.zip(gridModelMetrics.head().toSeq).toMap
 
-    modelMetricsFromGrid shouldEqual expectedMetrics
+    modelMetricsFromGrid.filter(!_._2.asInstanceOf[Double].isNaN) should contain theSameElementsAs expectedMetrics
+      .filter(!_._2.isNaN)
   }
 
   test("The first row returned by getGridModelsParams() method is the same as training params of the best model") {

diff --git a/py/tests/unit/with_runtime_sparkling/test_gridsearch.py b/py/tests/unit/with_runtime_sparkling/test_gridsearch.py
@@ -109,7 +109,7 @@ def testGetGridModelsMetrics(prostateDataset):
     grid.fit(prostateDataset)
     metrics = grid.getGridModelsMetrics()
     assert metrics.count() == 3
-    expectedCols = ['MOJO Model ID', 'RMSLE', 'Nobs', 'RMSE', 'MAE', 'MeanResidualDeviance', 'ScoringTime', 'MSE', 'R2']
+    expectedCols = ['MOJO Model ID', 'RMSLE', 'Nobs', 'RMSE', 'MAE', 'MeanResidualDeviance', 'ScoringTime', "Loglikelihood", 'MSE', 'R2', 'AIC']
     assert metrics.columns == expectedCols
     metrics.collect()  # try materializing
 

diff --git a/py/tests/unit/with_runtime_sparkling/test_mojo.py b/py/tests/unit/with_runtime_sparkling/test_mojo.py
@@ -19,6 +19,7 @@
 import shutil
 import unit_test_utils
 import os
+import math
 
 from pyspark.mllib.linalg import *
 from pyspark.sql.types import *
@@ -63,7 +64,7 @@ def testModelCategory(gbmModel):
 def testTrainingMetrics(gbmModel):
     metrics = gbmModel.getTrainingMetrics()
     assert metrics is not None
-    assert len(metrics) is 10
+    assert len(metrics) is 12
 
 
 def testFeatureTypes(gbmModel):
@@ -245,7 +246,8 @@ def compareMetricValues(metricsObject, metricsMap):
         for metric in metricsMap:
             metricValue = metricsMap[metric]
             objectValue = getattr(metricsObject, "get" + metric)()
-            assert(metricValue == objectValue)
+            if not math.isnan(metricValue) and not math.isnan(objectValue):
+                assert(metricValue == objectValue)
         assert metricsObject.getConfusionMatrix().count() > 0
         assert len(metricsObject.getConfusionMatrix().columns) > 0
         assert metricsObject.getGainsLiftTable().count() > 0