diff --git a/python/pyspark/ml/tests.py b/python/pyspark/ml/tests.py index a0c288a0b71a2023f4b28d7226a6e899f0b2b27f..68f5bc30ac57f854485c682dad3c30926b1bb313 100755 --- a/python/pyspark/ml/tests.py +++ b/python/pyspark/ml/tests.py @@ -390,6 +390,24 @@ class ParamTests(PySparkTestCase): self.assertEqual(model.getWindowSize(), 6) +class EvaluatorTests(SparkSessionTestCase): + + def test_java_params(self): + """ + This tests a bug fixed by SPARK-18274 which causes multiple copies + of a Params instance in Python to be linked to the same Java instance. + """ + evaluator = RegressionEvaluator(metricName="r2") + df = self.spark.createDataFrame([Row(label=1.0, prediction=1.1)]) + evaluator.evaluate(df) + self.assertEqual(evaluator._java_obj.getMetricName(), "r2") + evaluatorCopy = evaluator.copy({evaluator.metricName: "mae"}) + evaluator.evaluate(df) + evaluatorCopy.evaluate(df) + self.assertEqual(evaluator._java_obj.getMetricName(), "r2") + self.assertEqual(evaluatorCopy._java_obj.getMetricName(), "mae") + + class FeatureTests(SparkSessionTestCase): def test_binarizer(self): diff --git a/python/pyspark/ml/wrapper.py b/python/pyspark/ml/wrapper.py index 25c44b7533c7769292e1a9905a698f92901cbae8..13b75e99192215cd1f29db815adfd8bc92029fcc 100644 --- a/python/pyspark/ml/wrapper.py +++ b/python/pyspark/ml/wrapper.py @@ -71,6 +71,10 @@ class JavaParams(JavaWrapper, Params): __metaclass__ = ABCMeta + def __del__(self): + if SparkContext._active_spark_context: + SparkContext._active_spark_context._gateway.detach(self._java_obj) + def _make_java_param_pair(self, param, value): """ Makes a Java parm pair. @@ -180,6 +184,25 @@ class JavaParams(JavaWrapper, Params): % stage_name) return py_stage + def copy(self, extra=None): + """ + Creates a copy of this instance with the same uid and some + extra params. This implementation first calls Params.copy and + then make a copy of the companion Java pipeline component with + extra params. So both the Python wrapper and the Java pipeline + component get copied. + + :param extra: Extra parameters to copy to the new instance + :return: Copy of this instance + """ + if extra is None: + extra = dict() + that = super(JavaParams, self).copy(extra) + if self._java_obj is not None: + that._java_obj = self._java_obj.copy(self._empty_java_param_map()) + that._transfer_params_to_java() + return that + @inherit_doc class JavaEstimator(JavaParams, Estimator): @@ -256,21 +279,3 @@ class JavaModel(JavaTransformer, Model): super(JavaModel, self).__init__(java_model) if java_model is not None: self._resetUid(java_model.uid()) - - def copy(self, extra=None): - """ - Creates a copy of this instance with the same uid and some - extra params. This implementation first calls Params.copy and - then make a copy of the companion Java model with extra params. - So both the Python wrapper and the Java model get copied. - - :param extra: Extra parameters to copy to the new instance - :return: Copy of this instance - """ - if extra is None: - extra = dict() - that = super(JavaModel, self).copy(extra) - if self._java_obj is not None: - that._java_obj = self._java_obj.copy(self._empty_java_param_map()) - that._transfer_params_to_java() - return that