diff --git a/python/pyspark/ml/param/__init__.py b/python/pyspark/ml/param/__init__.py
index a1265294a1e9e..6b5f639496527 100644
--- a/python/pyspark/ml/param/__init__.py
+++ b/python/pyspark/ml/param/__init__.py
@@ -469,10 +469,11 @@ def _resetUid(self, newUid):
         Changes the uid of this instance. This updates both
         the stored uid and the parent uid of params and param maps.
         This is used by persistence (loading).
-        :param newUid: new uid to use
+        :param newUid: new uid to use, which is converted to unicode
         :return: same instance, but with the uid and Param.parent values
                  updated, including within param maps
         """
+        newUid = unicode(newUid)
         self.uid = newUid
         newDefaultParamMap = dict()
         newParamMap = dict()
diff --git a/python/pyspark/ml/tests.py b/python/pyspark/ml/tests.py
index 0b0ad2377fd7d..d6a0983daf83c 100644
--- a/python/pyspark/ml/tests.py
+++ b/python/pyspark/ml/tests.py
@@ -618,6 +618,8 @@ def test_linear_regression(self):
         lr_path = path + "/lr"
         lr.save(lr_path)
         lr2 = LinearRegression.load(lr_path)
+        self.assertEqual(lr.uid, lr2.uid)
+        self.assertEqual(type(lr.uid), type(lr2.uid))
         self.assertEqual(lr2.uid, lr2.maxIter.parent,
                          "Loaded LinearRegression instance uid (%s) did not match Param's uid (%s)"
                          % (lr2.uid, lr2.maxIter.parent))
diff --git a/python/pyspark/ml/util.py b/python/pyspark/ml/util.py
index 9dfcef0e40d67..841bfb47e1b9d 100644
--- a/python/pyspark/ml/util.py
+++ b/python/pyspark/ml/util.py
@@ -21,6 +21,7 @@
 
 if sys.version > '3':
     basestring = str
+    unicode = str
 
 from pyspark import SparkContext, since
 from pyspark.mllib.common import inherit_doc
@@ -67,10 +68,10 @@ def __repr__(self):
     @classmethod
     def _randomUID(cls):
         """
-        Generate a unique id for the object. The default implementation
+        Generate a unique unicode id for the object. The default implementation
         concatenates the class name, "_", and 12 random hex chars.
         """
-        return cls.__name__ + "_" + uuid.uuid4().hex[12:]
+        return unicode(cls.__name__ + "_" + uuid.uuid4().hex[12:])
 
 
 @inherit_doc
diff --git a/python/pyspark/ml/wrapper.py b/python/pyspark/ml/wrapper.py
index cd0e5b80d5559..e386c7095f63f 100644
--- a/python/pyspark/ml/wrapper.py
+++ b/python/pyspark/ml/wrapper.py
@@ -254,7 +254,7 @@ def __init__(self, java_model=None):
         """
         super(JavaModel, self).__init__(java_model)
         if java_model is not None:
-            self.uid = java_model.uid()
+            self._resetUid(java_model.uid())
 
     def copy(self, extra=None):
         """