[SPARK-14605][ML][PYTHON] Changed Python to use unicode UIDs for spark.ml Identifiable

## What changes were proposed in this pull request?

Python spark.ml Identifiable classes use UIDs of type str, but they should use unicode (in Python 2.x) to match Java. This could be a problem if someone created a class in Java with odd unicode characters, saved it, and loaded it in Python.

This PR: Use unicode everywhere in Python.

## How was this patch tested?

Updated persistence unit test to check uid type

Author: Joseph K. Bradley <joseph@databricks.com>

Closes #12368 from jkbradley/python-uid-unicode.

1 files changed, 3 insertions, 2 deletions

diff --git a/python/pyspark/ml/util.py b/python/pyspark/ml/util.py
index 9dfcef0e40..841bfb47e1 100644
--- a/python/pyspark/ml/util.py
+++ b/python/pyspark/ml/util.py
@@ -21,6 +21,7 @@ from functools import wraps
 
 if sys.version > '3':
     basestring = str
+    unicode = str
 
 from pyspark import SparkContext, since
 from pyspark.mllib.common import inherit_doc
@@ -67,10 +68,10 @@ class Identifiable(object):
     @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