Adding unicode / PY3 support for feature column vocab files.

PiperOrigin-RevId: 285862836
Change-Id: I2eec29c2300dfbc99f29b30b56e3e7dfea6d047e

tensorflow/python/feature_column/BUILD

@@ -99,6 +99,7 @@ filegroup(
         "testdata/embedding.ckpt.data-00000-of-00001",
         "testdata/embedding.ckpt.index",
         "testdata/embedding.ckpt.meta",
+        "testdata/unicode_vocabulary",
         "testdata/warriors_vocabulary.txt",
         "testdata/wire_vocabulary.txt",
     ],

tensorflow/python/feature_column/feature_column_v2.py

@@ -1665,7 +1665,7 @@ def categorical_column_with_vocabulary_file_v2(key,
     if not gfile.Exists(vocabulary_file):
       raise ValueError('vocabulary_file in {} does not exist.'.format(key))
 
-    with gfile.GFile(vocabulary_file) as f:
+    with gfile.GFile(vocabulary_file, mode='rb') as f:
       vocabulary_size = sum(1 for _ in f)
     logging.info(
         'vocabulary_size = %d in %s is inferred from the number of elements '

tensorflow/python/feature_column/feature_column_v2_test.py

@@ -3886,6 +3886,10 @@ class VocabularyFileCategoricalColumnTest(test.TestCase):
         'python/feature_column/testdata/wire_vocabulary.txt')
     self._wire_vocabulary_size = 3
 
+    # Contains unicode characters.
+    self._unicode_vocabulary_file_name = test.test_src_dir_path(
+        'python/feature_column/testdata/unicode_vocabulary')
+
   @test_util.run_deprecated_v1
   def test_defaults(self):
     column = fc.categorical_column_with_vocabulary_file(
@@ -3898,6 +3902,17 @@ class VocabularyFileCategoricalColumnTest(test.TestCase):
     }, column.parse_example_spec)
     self.assertTrue(column._is_v2_column)
 
+  @test_util.run_deprecated_v1
+  def test_defaults_unicode(self):
+    column = fc.categorical_column_with_vocabulary_file(
+        key='aaa', vocabulary_file=self._unicode_vocabulary_file_name)
+    self.assertEqual('aaa', column.name)
+    self.assertEqual('aaa', column.key)
+    self.assertEqual(165, column.num_buckets)
+    self.assertEqual({'aaa': parsing_ops.VarLenFeature(dtypes.string)},
+                     column.parse_example_spec)
+    self.assertTrue(column._is_v2_column)
+
   def test_key_should_be_string(self):
     with self.assertRaisesRegexp(ValueError, 'key must be a string.'):
       fc.categorical_column_with_vocabulary_file(

tensorflow/python/feature_column/testdata/unicode_vocabulary

@@ -0,0 +1,165 @@
+t
+/
+e
+o
+a
+s
+p
+i
+c
+n
+.
+r
+h
+m
+x
+l
+d
+w
+-
+u
+g
+b
+:
+2
+0
+1
+f
+%
+8
+3
+5
+k
+9
+4
+y
+7
+6
+v
+=
+_
+?
+A
+D
+j
+&
+F
+z
+E
+B
+S
+C
+q
+M
+L
+I
+R
+T
+N
+W
+P
+U
+G
+Z
+O
+V
+Y
+H
+J
+X
+Q
+K
++
+#
+,
+;
+~
+)
+@
+!
+|
+'
+(
+$
+*
+]
+[
+{
+}
+\
+ 
+^
+`
+"
+¸
+à
+Ð
+Ñ
+Ã
+€
+>
+<
+²
+°
+™
+±
+§
+„
+Ø
+‹
+ˆ
+â
+¾
+¹
+µ
+³
+Œ
+ì
+Ù
+º
+¡
+‚
+·
+©
+ 
+‡
+¼
+­
+ª
+<EFBFBD>
+ë
+å
+¿
+½
+´
+£
+—
+‰
+ê
+é
+è
+ç
+Û
+»
+¯
+¦
+¥
+¢
+ž
+›
+š
+˜
+”
+<EFBFBD>
+<EFBFBD>
+†
+ƒ
+í
+Ú
+Å
+¶
+®
+¨
+¤
+œ
+“