[SPARK-12006][ML][PYTHON] Fix GMM failure if initialModel is not None

If initial model passed to GMM is not empty it causes net.razorvine.pickle.PickleException. It can be fixed by converting initialModel.weights to list.

Author: zero323 <[email protected]>

Closes #10644 from zero323/SPARK-12006.

(cherry picked from commit 592f649)
Signed-off-by: Joseph K. Bradley <[email protected]>

Author: zero323

python/pyspark/mllib/clustering.py

@@ -255,7 +255,7 @@ def train(cls, rdd, k, convergenceTol=1e-3, maxIterations=100, seed=None, initia
             if initialModel.k != k:
                 raise Exception("Mismatched cluster count, initialModel.k = %s, however k = %s"
                                 % (initialModel.k, k))
-            initialModelWeights = initialModel.weights
+            initialModelWeights = list(initialModel.weights)
             initialModelMu = [initialModel.gaussians[i].mu for i in range(initialModel.k)]
             initialModelSigma = [initialModel.gaussians[i].sigma for i in range(initialModel.k)]
         weight, mu, sigma = callMLlibFunc("trainGaussianMixture", rdd.map(_convert_to_vector), k,

python/pyspark/mllib/tests.py

@@ -310,6 +310,18 @@ def test_gmm_deterministic(self):
         for c1, c2 in zip(clusters1.weights, clusters2.weights):
             self.assertEquals(round(c1, 7), round(c2, 7))
 
+    def test_gmm_with_initial_model(self):
+        from pyspark.mllib.clustering import GaussianMixture
+        data = self.sc.parallelize([
+            (-10, -5), (-9, -4), (10, 5), (9, 4)
+        ])
+
+        gmm1 = GaussianMixture.train(data, 2, convergenceTol=0.001,
+                                     maxIterations=10, seed=63)
+        gmm2 = GaussianMixture.train(data, 2, convergenceTol=0.001,
+                                     maxIterations=10, seed=63, initialModel=gmm1)
+        self.assertAlmostEqual((gmm1.weights - gmm2.weights).sum(), 0.0)
+
     def test_classification(self):
         from pyspark.mllib.classification import LogisticRegressionWithSGD, SVMWithSGD, NaiveBayes
         from pyspark.mllib.tree import DecisionTree, DecisionTreeModel, RandomForest,\