Removing underscore prefixes from hidden generated Python functions.

PiperOrigin-RevId: 187386941

23 files changed, 193 insertions, 199 deletions

diff --git a/tensorflow/python/kernel_tests/array_ops_test.py b/tensorflow/python/kernel_tests/array_ops_test.py
index 365cf72108..d35f62b186 100644
--- a/tensorflow/python/kernel_tests/array_ops_test.py
+++ b/tensorflow/python/kernel_tests/array_ops_test.py
@@ -1223,7 +1223,7 @@ class SnapshotOpTest(test_util.TensorFlowTestCase):
     for dtype in [dtypes.int32, dtypes.int64, dtypes.float32, dtypes.float64]:
       with self.test_session(use_gpu=True):
         x = constant_op.constant([0, 1, 2, 3], dtype=dtype)
-        y = gen_array_ops._snapshot(x)
+        y = gen_array_ops.snapshot(x)
         self.assertAllEqual(y.eval(), [0, 1, 2, 3])
 
 
diff --git a/tensorflow/python/kernel_tests/batchtospace_op_test.py b/tensorflow/python/kernel_tests/batchtospace_op_test.py
index 0c802476a0..6143cd3baa 100644
--- a/tensorflow/python/kernel_tests/batchtospace_op_test.py
+++ b/tensorflow/python/kernel_tests/batchtospace_op_test.py
@@ -44,7 +44,7 @@ class CppOpImpl(object):
 
   @staticmethod
   def batch_to_space(*args, **kwargs):
-    return gen_array_ops._batch_to_space(*args, **kwargs)
+    return gen_array_ops.batch_to_space(*args, **kwargs)
 
 
 class BatchToSpaceDepthToSpace(test.TestCase, PythonOpImpl):
diff --git a/tensorflow/python/kernel_tests/bcast_ops_test.py b/tensorflow/python/kernel_tests/bcast_ops_test.py
index 9e51234605..cb46fcb007 100644
--- a/tensorflow/python/kernel_tests/bcast_ops_test.py
+++ b/tensorflow/python/kernel_tests/bcast_ops_test.py
@@ -20,8 +20,8 @@ from __future__ import print_function
 
 from tensorflow.python.framework import constant_op
 from tensorflow.python.framework import dtypes
-from tensorflow.python.ops.gen_array_ops import _broadcast_args
 from tensorflow.python.ops.gen_array_ops import _broadcast_gradient_args
+from tensorflow.python.ops.gen_array_ops import broadcast_args
 from tensorflow.python.platform import test
 
 
@@ -29,7 +29,7 @@ class BcastOpsTest(test.TestCase):
 
   def _GetBroadcastShape(self, xs, ys):
     with self.test_session() as sess:
-      return sess.run(_broadcast_args(xs, ys))
+      return sess.run(broadcast_args(xs, ys))
 
   def _GetGradientArgs(self, xs, ys):
     with self.test_session() as sess:
diff --git a/tensorflow/python/kernel_tests/checkpoint_ops_test.py b/tensorflow/python/kernel_tests/checkpoint_ops_test.py
index a786d0a47e..7f147ba53a 100644
--- a/tensorflow/python/kernel_tests/checkpoint_ops_test.py
+++ b/tensorflow/python/kernel_tests/checkpoint_ops_test.py
@@ -50,7 +50,7 @@ class GenerateVocabRemappingTest(test.TestCase):
 
   def test_generate_remapping_with_no_vocab_changes(self):
     """Tests where vocab does not change at all."""
-    remapping, num_present = gen_checkpoint_ops._generate_vocab_remapping(
+    remapping, num_present = gen_checkpoint_ops.generate_vocab_remapping(
         new_vocab_file=self.old_vocab_file,
         old_vocab_file=self.old_vocab_file,
         num_new_vocab=3,
@@ -63,7 +63,7 @@ class GenerateVocabRemappingTest(test.TestCase):
 
   def test_generate_remapping_with_shifted_vocab(self):
     """Tests where vocab is the same, but shifted / ordered differently."""
-    remapping, num_present = gen_checkpoint_ops._generate_vocab_remapping(
+    remapping, num_present = gen_checkpoint_ops.generate_vocab_remapping(
         new_vocab_file=self.new_vocab_file,
         old_vocab_file=self.old_vocab_file,
         num_new_vocab=3,
@@ -76,7 +76,7 @@ class GenerateVocabRemappingTest(test.TestCase):
 
   def test_generate_remapping_with_offset(self):
     """Tests offset and num_new_vocab logic."""
-    remapping, num_present = gen_checkpoint_ops._generate_vocab_remapping(
+    remapping, num_present = gen_checkpoint_ops.generate_vocab_remapping(
         new_vocab_file=self.new_vocab_file,
         old_vocab_file=self.old_vocab_file,
         num_new_vocab=1,
@@ -89,7 +89,7 @@ class GenerateVocabRemappingTest(test.TestCase):
 
   def test_generate_remapping_with_old_vocab_size(self):
     """Tests where old_vocab_size is specified."""
-    remapping, num_present = gen_checkpoint_ops._generate_vocab_remapping(
+    remapping, num_present = gen_checkpoint_ops.generate_vocab_remapping(
         new_vocab_file=self.new_vocab_file,
         old_vocab_file=self.old_vocab_file,
         num_new_vocab=3,
@@ -132,7 +132,7 @@ class LoadAndRemapMatrixTest(test.TestCase):
 
     # No column remapping, new weight matrix has second row, then first row.
     row_remapping = [1, 0]
-    remapped_matrix = gen_checkpoint_ops._load_and_remap_matrix(
+    remapped_matrix = gen_checkpoint_ops.load_and_remap_matrix(
         ckpt_path=[self.bundle_file],
         old_tensor_name=self.old_tensor_name,
         row_remapping=row_remapping,
@@ -147,7 +147,7 @@ class LoadAndRemapMatrixTest(test.TestCase):
     # No row remapping, new weight matrix has third col, then first col.
     row_remapping = list(range(self.old_num_rows))
     col_remapping = [2, 0]
-    remapped_matrix = gen_checkpoint_ops._load_and_remap_matrix(
+    remapped_matrix = gen_checkpoint_ops.load_and_remap_matrix(
         ckpt_path=[self.bundle_file],
         old_tensor_name=self.old_tensor_name,
         row_remapping=row_remapping,
@@ -162,7 +162,7 @@ class LoadAndRemapMatrixTest(test.TestCase):
     # Both row and column remappings.
     row_remapping = [1, 0, 4]
     col_remapping = [1, 15]
-    remapped_matrix = gen_checkpoint_ops._load_and_remap_matrix(
+    remapped_matrix = gen_checkpoint_ops.load_and_remap_matrix(
         ckpt_path=[self.bundle_file],
         old_tensor_name=self.old_tensor_name,
         row_remapping=row_remapping,
@@ -177,7 +177,7 @@ class LoadAndRemapMatrixTest(test.TestCase):
   def test_load_and_remap_with_init(self):
     """Tests the op's load and remap where there are missing entries."""
     init_val = 42
-    remapped_matrix = gen_checkpoint_ops._load_and_remap_matrix(
+    remapped_matrix = gen_checkpoint_ops.load_and_remap_matrix(
         ckpt_path=[self.bundle_file],
         old_tensor_name=self.old_tensor_name,
         row_remapping=[2, -1, 0],
@@ -196,7 +196,7 @@ class LoadAndRemapMatrixTest(test.TestCase):
     """Tests when all the rows are missing and need to be initialized."""
     num_rows = 7
     initializing_values = [42] * num_rows * self.old_num_cols
-    remapped_matrix = gen_checkpoint_ops._load_and_remap_matrix(
+    remapped_matrix = gen_checkpoint_ops.load_and_remap_matrix(
         ckpt_path=[self.bundle_file],
         old_tensor_name=self.old_tensor_name,
         row_remapping=[-1] * num_rows,
@@ -214,7 +214,7 @@ class LoadAndRemapMatrixTest(test.TestCase):
     num_rows = 7
     num_cols = 4
     initializing_values = [42] * num_rows * num_cols
-    remapped_matrix = gen_checkpoint_ops._load_and_remap_matrix(
+    remapped_matrix = gen_checkpoint_ops.load_and_remap_matrix(
         ckpt_path=[self.bundle_file],
         old_tensor_name=self.old_tensor_name,
         row_remapping=[-1] * num_rows,
@@ -235,7 +235,7 @@ class LoadAndRemapMatrixTest(test.TestCase):
     invalid_remapping = [1, 0, 0, 0, 1, 2]
 
     # Invalid row remapping.
-    remapped_matrix = gen_checkpoint_ops._load_and_remap_matrix(
+    remapped_matrix = gen_checkpoint_ops.load_and_remap_matrix(
         ckpt_path=[self.bundle_file],
         old_tensor_name=self.old_tensor_name,
         row_remapping=invalid_remapping,
@@ -247,7 +247,7 @@ class LoadAndRemapMatrixTest(test.TestCase):
       remapped_matrix.eval()
 
     # Invalid column remapping.
-    remapped_matrix = gen_checkpoint_ops._load_and_remap_matrix(
+    remapped_matrix = gen_checkpoint_ops.load_and_remap_matrix(
         ckpt_path=[self.bundle_file],
         old_tensor_name=self.old_tensor_name,
         row_remapping=list(range(self.old_num_rows)),
@@ -260,7 +260,7 @@ class LoadAndRemapMatrixTest(test.TestCase):
 
   def test_load_and_remap_incorrect_initializing_values(self):
     """Tests that errors are raised with incorrect number of init values."""
-    remapped_matrix = gen_checkpoint_ops._load_and_remap_matrix(
+    remapped_matrix = gen_checkpoint_ops.load_and_remap_matrix(
         ckpt_path=[self.bundle_file],
         old_tensor_name=self.old_tensor_name,
         row_remapping=[2, -1, 0],
@@ -275,7 +275,7 @@ class LoadAndRemapMatrixTest(test.TestCase):
     with self.test_session(), self.assertRaises(errors.InvalidArgumentError):
       remapped_matrix.eval()
 
-    remapped_matrix = gen_checkpoint_ops._load_and_remap_matrix(
+    remapped_matrix = gen_checkpoint_ops.load_and_remap_matrix(
         ckpt_path=[self.bundle_file],
         old_tensor_name=self.old_tensor_name,
         row_remapping=[2, -1, 0],
@@ -314,7 +314,7 @@ class LoadAndRemapMatrixWithMaxRowsTest(test.TestCase):
       num_rows, num_cols = np_value.shape
 
       # Tests loading the entire tensor (except reversed).
-      remapped_matrix = gen_checkpoint_ops._load_and_remap_matrix(
+      remapped_matrix = gen_checkpoint_ops.load_and_remap_matrix(
           ckpt_path=ckpt_path,
           old_tensor_name=old_tensor_name,
           # Simply reverses the rows of the matrix.
@@ -332,7 +332,7 @@ class LoadAndRemapMatrixWithMaxRowsTest(test.TestCase):
       self.assertGreater(num_rows, 2)
       prefix_rows = 2
       suffix_rows = 3
-      remapped_matrix = gen_checkpoint_ops._load_and_remap_matrix(
+      remapped_matrix = gen_checkpoint_ops.load_and_remap_matrix(
           ckpt_path=ckpt_path,
           old_tensor_name=old_tensor_name,
           # Reverses the rows of the matrix, then prepends and appends
@@ -353,7 +353,7 @@ class LoadAndRemapMatrixWithMaxRowsTest(test.TestCase):
       # Tests when everything is taken from initializing_values.
       new_rows = 7
       initializing_values = [42] * new_rows * num_cols
-      remapped_matrix = gen_checkpoint_ops._load_and_remap_matrix(
+      remapped_matrix = gen_checkpoint_ops.load_and_remap_matrix(
           ckpt_path=ckpt_path,
           old_tensor_name=old_tensor_name,
           # Nothing is loaded from the old tensor.
diff --git a/tensorflow/python/kernel_tests/concat_op_test.py b/tensorflow/python/kernel_tests/concat_op_test.py
index 127bc6bb20..81c6a4aa6e 100644
--- a/tensorflow/python/kernel_tests/concat_op_test.py
+++ b/tensorflow/python/kernel_tests/concat_op_test.py
@@ -526,7 +526,7 @@ class ConcatOpTest(test.TestCase):
     with self.test_session(use_gpu=True):
       t1 = []
       t2 = []
-      output = gen_array_ops._concat_v2([t1, t2], 0).eval()
+      output = gen_array_ops.concat_v2([t1, t2], 0).eval()
       self.assertFalse(output)  # Checks that output is empty
 
   def testConcatInvalidAxis(self):
@@ -534,20 +534,20 @@ class ConcatOpTest(test.TestCase):
       with self.test_session(use_gpu=True):
         t1 = [1]
         t2 = [2]
-        gen_array_ops._concat_v2([t1, t2], 1).eval()
+        gen_array_ops.concat_v2([t1, t2], 1).eval()
 
   def testConcatNegativeAxis(self):
     with self.test_session(use_gpu=True):
       t1 = [[1, 2, 3], [4, 5, 6]]
       t2 = [[7, 8, 9], [10, 11, 12]]
 
-      c = gen_array_ops._concat_v2([t1, t2], -2)
+      c = gen_array_ops.concat_v2([t1, t2], -2)
       self.assertEqual([4, 3], c.get_shape().as_list())
       output = c.eval()
       self.assertAllEqual([[1, 2, 3], [4, 5, 6], [7, 8, 9], [10, 11, 12]],
                           output)
 
-      c = gen_array_ops._concat_v2([t1, t2], -1)
+      c = gen_array_ops.concat_v2([t1, t2], -1)
       self.assertEqual([2, 6], c.get_shape().as_list())
       output = c.eval()
       self.assertAllEqual([[1, 2, 3, 7, 8, 9], [4, 5, 6, 10, 11, 12]], output)
@@ -615,7 +615,7 @@ class ConcatOffsetTest(test.TestCase):
       s0 = constant_op.constant([2, 3, 5], dtypes.int32)
       s1 = constant_op.constant([2, 7, 5], dtypes.int32)
       s2 = constant_op.constant([2, 20, 5], dtypes.int32)
-      off = gen_array_ops._concat_offset(cdim, [s0, s1, s2])
+      off = gen_array_ops.concat_offset(cdim, [s0, s1, s2])
       ans = sess.run(off)
       self.assertAllEqual(ans, [[0, 0, 0], [0, 3, 0], [0, 10, 0]])
 
@@ -624,7 +624,7 @@ class ConcatOffsetTest(test.TestCase):
       cdim = constant_op.constant(1, dtypes.int32)
       s0 = constant_op.constant([[2, 3, 5]], dtypes.int32)
       s1 = constant_op.constant([[2, 7, 5]], dtypes.int32)
-      off = gen_array_ops._concat_offset(cdim, [s0, s1])
+      off = gen_array_ops.concat_offset(cdim, [s0, s1])
       with self.assertRaisesRegexp(errors_impl.InvalidArgumentError,
                                    r"should be a vector"):
         sess.run(off)
@@ -634,7 +634,7 @@ class ConcatOffsetTest(test.TestCase):
       cdim = constant_op.constant(4, dtypes.int32)
       s0 = constant_op.constant([2, 3, 5], dtypes.int32)
       s1 = constant_op.constant([2, 7, 5], dtypes.int32)
-      off = gen_array_ops._concat_offset(cdim, [s0, s1])
+      off = gen_array_ops.concat_offset(cdim, [s0, s1])
       with self.assertRaisesRegexp(errors_impl.InvalidArgumentError,
                                    r"Concat dim is out of range: 4 vs. 3"):
         sess.run(off)
@@ -644,7 +644,7 @@ class ConcatOffsetTest(test.TestCase):
       cdim = constant_op.constant(1, dtypes.int32)
       s0 = constant_op.constant([2, 3, 5], dtypes.int32)
       s1 = constant_op.constant([2, 7, 5, 10], dtypes.int32)
-      off = gen_array_ops._concat_offset(cdim, [s0, s1])
+      off = gen_array_ops.concat_offset(cdim, [s0, s1])
       with self.assertRaisesRegexp(errors_impl.InvalidArgumentError,
                                    r"should contain 3 elem"):
         sess.run(off)
@@ -654,7 +654,7 @@ class ConcatOffsetTest(test.TestCase):
       cdim = constant_op.constant(1, dtypes.int32)
       s0 = constant_op.constant([2, 3, 5], dtypes.int32)
       s1 = constant_op.constant([2, 7, 10], dtypes.int32)
-      off = gen_array_ops._concat_offset(cdim, [s0, s1])
+      off = gen_array_ops.concat_offset(cdim, [s0, s1])
       with self.assertRaisesRegexp(
           errors_impl.InvalidArgumentError,
           r"All dimensions except 1 must match. Input 1 has shape \[2 7 10\] "
@@ -667,7 +667,7 @@ class ConcatOffsetTest(test.TestCase):
       s0 = constant_op.constant([2, 3, 5], dtypes.int32)
       s1 = constant_op.constant([2, 7, 5], dtypes.int32)
       s2 = constant_op.constant([2, 20, 5], dtypes.int32)
-      off = gen_array_ops._concat_offset(cdim, [s0, s1, s2])
+      off = gen_array_ops.concat_offset(cdim, [s0, s1, s2])
       ans = sess.run(off)
       self.assertAllEqual(ans, [[0, 0, 0], [0, 3, 0], [0, 10, 0]])
 
@@ -675,7 +675,7 @@ class ConcatOffsetTest(test.TestCase):
       s0 = constant_op.constant([2, 3, 5], dtypes.int32)
       s1 = constant_op.constant([1, 3, 5], dtypes.int32)
       s2 = constant_op.constant([3, 3, 5], dtypes.int32)
-      off = gen_array_ops._concat_offset(cdim, [s0, s1, s2])
+      off = gen_array_ops.concat_offset(cdim, [s0, s1, s2])
       ans = sess.run(off)
       self.assertAllEqual(ans, [[0, 0, 0], [2, 0, 0], [3, 0, 0]])
 
diff --git a/tensorflow/python/kernel_tests/control_flow_ops_py_test.py b/tensorflow/python/kernel_tests/control_flow_ops_py_test.py
index 58f38650eb..b429fa5c42 100644
--- a/tensorflow/python/kernel_tests/control_flow_ops_py_test.py
+++ b/tensorflow/python/kernel_tests/control_flow_ops_py_test.py
@@ -552,7 +552,7 @@ class ControlFlowTest(test.TestCase):
 
   def testCondRef(self):
     with self.test_session():
-      x = gen_state_ops._variable(
+      x = gen_state_ops.variable(
           shape=[1],
           dtype=dtypes.float32,
           name="x",
@@ -580,7 +580,7 @@ class ControlFlowTest(test.TestCase):
 
   def testUninitializedRefIdentity(self):
     with self.test_session() as sess:
-      v = gen_state_ops._variable(
+      v = gen_state_ops.variable(
           shape=[1],
           dtype=dtypes.float32,
           name="v",
@@ -1620,7 +1620,7 @@ class ControlFlowTest(test.TestCase):
 
   def testWhileStack_1(self):
     with self.test_session():
-      s = gen_data_flow_ops._stack_v2(-1, dtypes.int32, stack_name="foo")
+      s = gen_data_flow_ops.stack_v2(-1, dtypes.int32, stack_name="foo")
       i = constant_op.constant(0)
 
       def c(i):
@@ -1629,7 +1629,7 @@ class ControlFlowTest(test.TestCase):
       def b(i):
         ni = math_ops.add(i, 1)
         ni = control_flow_ops.with_dependencies(
-            [gen_data_flow_ops._stack_push_v2(s, i)], ni)
+            [gen_data_flow_ops.stack_push_v2(s, i)], ni)
         return ni
 
       r = control_flow_ops.while_loop(c, b, [i], parallel_iterations=1)
@@ -1641,7 +1641,7 @@ class ControlFlowTest(test.TestCase):
 
       def b1(i, x):
         ni = math_ops.subtract(i, 1)
-        nx = x + gen_data_flow_ops._stack_pop_v2(s, dtypes.int32)
+        nx = x + gen_data_flow_ops.stack_pop_v2(s, dtypes.int32)
         return [ni, nx]
 
       _, rx = control_flow_ops.while_loop(
diff --git a/tensorflow/python/kernel_tests/cwise_ops_test.py b/tensorflow/python/kernel_tests/cwise_ops_test.py
index 0d9b46c30d..8db0bb6f0d 100644
--- a/tensorflow/python/kernel_tests/cwise_ops_test.py
+++ b/tensorflow/python/kernel_tests/cwise_ops_test.py
@@ -495,11 +495,11 @@ class UnaryOpTest(test.TestCase):
     dtype_tols = [(np.float32, 5e-4), (np.float64, 1e-6), (np.complex64, 5e-4),
                   (np.complex128, 1e-6)]
     op_range = [
-        (gen_math_ops._reciprocal_grad, [-2, 2]),
-        (gen_math_ops._rsqrt_grad, [0.1, 3]),
-        (gen_math_ops._sigmoid_grad, [-2, 2]),
-        (gen_math_ops._sqrt_grad, [0.1, 3]),
-        (gen_math_ops._tanh_grad, [-2, 2]),
+        (gen_math_ops.reciprocal_grad, [-2, 2]),
+        (gen_math_ops.rsqrt_grad, [0.1, 3]),
+        (gen_math_ops.sigmoid_grad, [-2, 2]),
+        (gen_math_ops.sqrt_grad, [0.1, 3]),
+        (gen_math_ops.tanh_grad, [-2, 2]),
     ]
 
     def rand(dtype):
diff --git a/tensorflow/python/kernel_tests/determinant_op_test.py b/tensorflow/python/kernel_tests/determinant_op_test.py
index 222038b22e..a52b2c0dc3 100644
--- a/tensorflow/python/kernel_tests/determinant_op_test.py
+++ b/tensorflow/python/kernel_tests/determinant_op_test.py
@@ -65,7 +65,7 @@ class DeterminantOpTest(test.TestCase):
       self._compareDeterminantBase(matrix_x,
                                    linalg_ops.matrix_determinant(matrix_x))
       self._compareLogDeterminantBase(
-          matrix_x, gen_linalg_ops._log_matrix_determinant(matrix_x))
+          matrix_x, gen_linalg_ops.log_matrix_determinant(matrix_x))
 
   def testBasic(self):
     # 2x2 matrices
diff --git a/tensorflow/python/kernel_tests/fractional_avg_pool_op_test.py b/tensorflow/python/kernel_tests/fractional_avg_pool_op_test.py
index feec9934e4..faac7d8365 100644
--- a/tensorflow/python/kernel_tests/fractional_avg_pool_op_test.py
+++ b/tensorflow/python/kernel_tests/fractional_avg_pool_op_test.py
@@ -347,7 +347,7 @@ class FractionalAvgPoolGradTest(test.TestCase):
 
   Two types of tests for FractionalAvgPoolGrad.
   1) Test fractional_avg_pool_grad() directly.
-    This type of test relies on gen_nn_ops._avg_pool_grad() returns the
+    This type of test relies on gen_nn_ops.avg_pool_grad() returns the
   correct result. For example:
     * input_tensor_shape = (1, 10, 10, 1)
     * window_size = (1, 2, 2, 1)
@@ -404,13 +404,13 @@ class FractionalAvgPoolGradTest(test.TestCase):
                 num_elements *= dim_size
               output_backprop = (self._PRNG.rand(num_elements) *
                                  1000).reshape(output_data.shape)
-              input_backprop_tensor = gen_nn_ops._avg_pool_grad(
+              input_backprop_tensor = gen_nn_ops.avg_pool_grad(
                   input_tensor.get_shape(), output_backprop, window_size,
                   stride_size, padding)
               input_backprop = input_backprop_tensor.eval()
               row_seq = list(range(0, num_rows + 1, row_window_size))
               col_seq = list(range(0, num_cols + 1, col_window_size))
-              fap_input_backprop_tensor = gen_nn_ops._fractional_avg_pool_grad(
+              fap_input_backprop_tensor = gen_nn_ops.fractional_avg_pool_grad(
                   input_tensor.get_shape(),
                   output_backprop,
                   row_seq,
@@ -443,7 +443,7 @@ class FractionalAvgPoolGradTest(test.TestCase):
                 num_elements *= dim_size
               output_backprop = (self._PRNG.rand(num_elements) *
                                  1000).reshape(output_data.shape)
-              input_backprop_tensor = gen_nn_ops._avg_pool_grad(
+              input_backprop_tensor = gen_nn_ops.avg_pool_grad(
                   input_tensor.get_shape(), output_backprop, window_size,
                   stride_size, padding)
               input_backprop = input_backprop_tensor.eval()
@@ -451,7 +451,7 @@ class FractionalAvgPoolGradTest(test.TestCase):
               col_seq = list(range(0, num_cols, col_window_size - 1))
               row_seq[-1] += 1
               col_seq[-1] += 1
-              fap_input_backprop_tensor = gen_nn_ops._fractional_avg_pool_grad(
+              fap_input_backprop_tensor = gen_nn_ops.fractional_avg_pool_grad(
                   input_tensor.get_shape(),
                   output_backprop,
                   row_seq,
diff --git a/tensorflow/python/kernel_tests/fractional_max_pool_op_test.py b/tensorflow/python/kernel_tests/fractional_max_pool_op_test.py
index 5983ae7759..6477c9ebc4 100644
--- a/tensorflow/python/kernel_tests/fractional_max_pool_op_test.py
+++ b/tensorflow/python/kernel_tests/fractional_max_pool_op_test.py
@@ -318,7 +318,7 @@ class FractionalMaxPoolGradTest(test.TestCase):
 
   Two types of tests for FractionalMaxPoolGrad.
   1) Test fractional_max_pool_grad() directly.
-    This type of test relies on gen_nn_ops._max_pool_grad() returns the correct
+    This type of test relies on gen_nn_ops.max_pool_grad() returns the correct
   result. For example:
     * input_tensor_shape = (1, 10, 10, 1)
     * window_size = (1, 2, 2, 1)
@@ -384,16 +384,13 @@ class FractionalMaxPoolGradTest(test.TestCase):
                                               stride_size, padding)
               output_data = output_tensor.eval()
               output_backprop = self._PRNG.randint(100, size=output_data.shape)
-              input_backprop_tensor = gen_nn_ops._max_pool_grad(input_tensor,
-                                                                output_tensor,
-                                                                output_backprop,
-                                                                window_size,
-                                                                stride_size,
-                                                                padding)
+              input_backprop_tensor = gen_nn_ops.max_pool_grad(
+                  input_tensor, output_tensor, output_backprop, window_size,
+                  stride_size, padding)
               input_backprop = input_backprop_tensor.eval()
               row_seq = list(range(0, num_rows + 1, row_window_size))
               col_seq = list(range(0, num_cols + 1, col_window_size))
-              fmp_input_backprop_tensor = gen_nn_ops._fractional_max_pool_grad(
+              fmp_input_backprop_tensor = gen_nn_ops.fractional_max_pool_grad(
                   input_tensor,
                   output_tensor,
                   output_backprop,
@@ -422,18 +419,15 @@ class FractionalMaxPoolGradTest(test.TestCase):
                                               stride_size, padding)
               output_data = output_tensor.eval()
               output_backprop = self._PRNG.randint(100, size=output_data.shape)
-              input_backprop_tensor = gen_nn_ops._max_pool_grad(input_tensor,
-                                                                output_tensor,
-                                                                output_backprop,
-                                                                window_size,
-                                                                stride_size,
-                                                                padding)
+              input_backprop_tensor = gen_nn_ops.max_pool_grad(
+                  input_tensor, output_tensor, output_backprop, window_size,
+                  stride_size, padding)
               input_backprop = input_backprop_tensor.eval()
               row_seq = list(range(0, num_rows, row_window_size - 1))
               col_seq = list(range(0, num_cols, col_window_size - 1))
               row_seq[-1] += 1
               col_seq[-1] += 1
-              fmp_input_backprop_tensor = gen_nn_ops._fractional_max_pool_grad(
+              fmp_input_backprop_tensor = gen_nn_ops.fractional_max_pool_grad(
                   input_tensor,
                   output_tensor,
                   output_backprop,
@@ -591,7 +585,7 @@ class FractionalMaxPoolGradTest(test.TestCase):
       output_tensor = constant_op.constant(
           output_data_not_overlapping, shape=output_size)
       grad = constant_op.constant(output_backprop, shape=output_size)
-      r = gen_nn_ops._fractional_max_pool_grad(
+      r = gen_nn_ops.fractional_max_pool_grad(
           input_tensor,
           output_tensor,
           grad,
@@ -606,7 +600,7 @@ class FractionalMaxPoolGradTest(test.TestCase):
       # Test when overlapping is True
       output_tensor = constant_op.constant(
           output_data_overlapping, shape=output_size)
-      r = gen_nn_ops._fractional_max_pool_grad(
+      r = gen_nn_ops.fractional_max_pool_grad(
           input_tensor, output_tensor, grad, row_seq, col_seq, overlapping=True)
       input_backprop_overlapping = r.eval()
       self.assertShapeEqual(
diff --git a/tensorflow/python/kernel_tests/matrix_exponential_op_test.py b/tensorflow/python/kernel_tests/matrix_exponential_op_test.py
index 6203a412d7..a0c66c77d8 100644
--- a/tensorflow/python/kernel_tests/matrix_exponential_op_test.py
+++ b/tensorflow/python/kernel_tests/matrix_exponential_op_test.py
@@ -48,7 +48,7 @@ class ExponentialOpTest(test.TestCase):
   def _verifyExponential(self, x, np_type):
     inp = x.astype(np_type)
     with self.test_session(use_gpu=True):
-      tf_ans = gen_linalg_ops._matrix_exponential(inp)
+      tf_ans = gen_linalg_ops.matrix_exponential(inp)
       if x.size == 0:
         np_ans = np.empty(x.shape, dtype=np_type)
       else:
@@ -116,13 +116,13 @@ class ExponentialOpTest(test.TestCase):
     # When the exponential of a non-square matrix is attempted we should return
     # an error
     with self.assertRaises(ValueError):
-      gen_linalg_ops._matrix_exponential(np.array([[1., 2., 3.], [3., 4., 5.]]))
+      gen_linalg_ops.matrix_exponential(np.array([[1., 2., 3.], [3., 4., 5.]]))
 
   def testWrongDimensions(self):
     # The input to the exponential should be at least a 2-dimensional tensor.
     tensor3 = constant_op.constant([1., 2.])
     with self.assertRaises(ValueError):
-      gen_linalg_ops._matrix_exponential(tensor3)
+      gen_linalg_ops.matrix_exponential(tensor3)
 
   def testEmpty(self):
     self._verifyExponentialReal(np.empty([0, 2, 2]))
@@ -143,8 +143,8 @@ class ExponentialOpTest(test.TestCase):
     with self.test_session(use_gpu=True) as sess:
       matrix1 = random_ops.random_normal([5, 5], seed=42)
       matrix2 = random_ops.random_normal([5, 5], seed=42)
-      expm1 = gen_linalg_ops._matrix_exponential(matrix1)
-      expm2 = gen_linalg_ops._matrix_exponential(matrix2)
+      expm1 = gen_linalg_ops.matrix_exponential(matrix1)
+      expm2 = gen_linalg_ops.matrix_exponential(matrix2)
       expm = sess.run([expm1, expm2])
       self.assertAllEqual(expm[0], expm[1])
 
@@ -180,7 +180,7 @@ class MatrixExponentialBenchmark(test.Benchmark):
           session.Session() as sess, \
           ops.device("/cpu:0"):
         matrix = self._GenerateMatrix(shape)
-        expm = gen_linalg_ops._matrix_exponential(matrix)
+        expm = gen_linalg_ops.matrix_exponential(matrix)
         variables.global_variables_initializer().run()
         self.run_op_benchmark(
             sess,
diff --git a/tensorflow/python/kernel_tests/matrix_logarithm_op_test.py b/tensorflow/python/kernel_tests/matrix_logarithm_op_test.py
index 18ed59828c..24edc4f59f 100644
--- a/tensorflow/python/kernel_tests/matrix_logarithm_op_test.py
+++ b/tensorflow/python/kernel_tests/matrix_logarithm_op_test.py
@@ -39,8 +39,8 @@ class LogarithmOpTest(test.TestCase):
     inp = x.astype(np_type)
     with self.test_session(use_gpu=True):
       # Verify that expm(logm(A)) == A.
-      tf_ans = gen_linalg_ops._matrix_exponential(
-          gen_linalg_ops._matrix_logarithm(inp))
+      tf_ans = gen_linalg_ops.matrix_exponential(
+          gen_linalg_ops.matrix_logarithm(inp))
       out = tf_ans.eval()
       self.assertAllClose(inp, out, rtol=1e-4, atol=1e-3)
 
@@ -85,14 +85,14 @@ class LogarithmOpTest(test.TestCase):
     # When the logarithm of a non-square matrix is attempted we should return
     # an error
     with self.assertRaises(ValueError):
-      gen_linalg_ops._matrix_logarithm(
+      gen_linalg_ops.matrix_logarithm(
           np.array([[1., 2., 3.], [3., 4., 5.]], dtype=np.complex64))
 
   def testWrongDimensions(self):
     # The input to the logarithm should be at least a 2-dimensional tensor.
     tensor3 = constant_op.constant([1., 2.], dtype=dtypes.complex64)
     with self.assertRaises(ValueError):
-      gen_linalg_ops._matrix_logarithm(tensor3)
+      gen_linalg_ops.matrix_logarithm(tensor3)
 
   def testEmpty(self):
     self._verifyLogarithmComplex(np.empty([0, 2, 2], dtype=np.complex64))
@@ -115,8 +115,8 @@ class LogarithmOpTest(test.TestCase):
           random_ops.random_normal([5, 5], seed=42), dtypes.complex64)
       matrix2 = math_ops.cast(
           random_ops.random_normal([5, 5], seed=42), dtypes.complex64)
-      logm1 = gen_linalg_ops._matrix_logarithm(matrix1)
-      logm2 = gen_linalg_ops._matrix_logarithm(matrix2)
+      logm1 = gen_linalg_ops.matrix_logarithm(matrix1)
+      logm2 = gen_linalg_ops.matrix_logarithm(matrix2)
       logm = sess.run([logm1, logm2])
       self.assertAllEqual(logm[0], logm[1])
 
@@ -152,7 +152,7 @@ class MatrixLogarithmBenchmark(test.Benchmark):
           session.Session() as sess, \
           ops.device("/cpu:0"):
         matrix = self._GenerateMatrix(shape)
-        logm = gen_linalg_ops._matrix_logarithm(matrix)
+        logm = gen_linalg_ops.matrix_logarithm(matrix)
         variables.global_variables_initializer().run()
         self.run_op_benchmark(
             sess,
diff --git a/tensorflow/python/kernel_tests/pooling_ops_test.py b/tensorflow/python/kernel_tests/pooling_ops_test.py
index 4466beeec9..a0ac355b60 100644
--- a/tensorflow/python/kernel_tests/pooling_ops_test.py
+++ b/tensorflow/python/kernel_tests/pooling_ops_test.py
@@ -405,7 +405,7 @@ class PoolingTest(test.TestCase):
 
     for v2 in [True, False]:
       self._VerifyValues(
-          gen_nn_ops._max_pool_v2,
+          gen_nn_ops.max_pool_v2,
           input_sizes=[1, 3, 3, 3],
           ksize=[1, 2, 2, 1],
           strides=[1, 2, 2, 1],
@@ -427,7 +427,7 @@ class PoolingTest(test.TestCase):
 
     for v2 in [True, False]:
       self._VerifyValues(
-          gen_nn_ops._max_pool_v2,
+          gen_nn_ops.max_pool_v2,
           input_sizes=[1, 2, 3, 3],
           ksize=[1, 2, 2, 1],
           strides=[1, 2, 2, 1],
@@ -456,7 +456,7 @@ class PoolingTest(test.TestCase):
 
     for v2 in [True, False]:
       self._VerifyValues(
-          gen_nn_ops._max_pool_v2,
+          gen_nn_ops.max_pool_v2,
           input_sizes=[1, 2, 2, 1],
           ksize=[1, 1, 2, 1],
           strides=[1, 1, 1, 1],
@@ -485,7 +485,7 @@ class PoolingTest(test.TestCase):
 
     for v2 in [True, False]:
       self._VerifyValues(
-          gen_nn_ops._max_pool_v2,
+          gen_nn_ops.max_pool_v2,
           input_sizes=[1, 4, 4, 1],
           ksize=[1, 2, 2, 1],
           strides=[1, 1, 2, 1],
@@ -494,7 +494,7 @@ class PoolingTest(test.TestCase):
           use_gpu=use_gpu,
           v2=v2)
       self._VerifyValues(
-          gen_nn_ops._max_pool_v2,
+          gen_nn_ops.max_pool_v2,
           input_sizes=[1, 4, 4, 1],
           ksize=[1, 2, 2, 1],
           strides=[1, 2, 1, 1],
@@ -519,7 +519,7 @@ class PoolingTest(test.TestCase):
 
     for v2 in [True, False]:
       self._VerifyValues(
-          gen_nn_ops._max_pool_v2,
+          gen_nn_ops.max_pool_v2,
           input_sizes=[1, 4, 4, 4],
           ksize=[1, 2, 2, 1],
           strides=[1, 2, 2, 1],
@@ -554,7 +554,7 @@ class PoolingTest(test.TestCase):
 
     for v2 in [True, False]:
       self._VerifyValues(
-          gen_nn_ops._max_pool_v2,
+          gen_nn_ops.max_pool_v2,
           input_sizes=[1, 8, 8, 8],
           ksize=[1, 3, 3, 1],
           strides=[1, 2, 2, 1],
@@ -565,7 +565,7 @@ class PoolingTest(test.TestCase):
 
   def _testMaxPoolEmptyInput(self, use_gpu):
     self._VerifyValues(
-        gen_nn_ops._max_pool_v2,
+        gen_nn_ops.max_pool_v2,
         input_sizes=[0, 8, 8, 8],
         ksize=[1, 3, 3, 1],
         strides=[1, 2, 2, 1],
@@ -600,7 +600,7 @@ class PoolingTest(test.TestCase):
 
     for v2 in [True, False]:
       self._VerifyValues(
-          gen_nn_ops._max_pool_v2,
+          gen_nn_ops.max_pool_v2,
           input_sizes=[1, 1, 1, 10],
           ksize=[1, 1, 1, 2],
           strides=[1, 1, 1, 2],
@@ -626,7 +626,7 @@ class PoolingTest(test.TestCase):
 
     for v2 in [True, False]:
       self._VerifyValues(
-          gen_nn_ops._max_pool_v2,
+          gen_nn_ops.max_pool_v2,
           input_sizes=[1, 2, 2, 6],
           ksize=[1, 1, 1, 3],
           strides=[1, 1, 1, 3],
@@ -648,7 +648,7 @@ class PoolingTest(test.TestCase):
 
       for v2 in [True, False]:
         self._VerifyValues(
-            gen_nn_ops._max_pool_v2,
+            gen_nn_ops.max_pool_v2,
             input_sizes=[1, 7, 7, 1],
             ksize=[1, 2, 2, 1],
             strides=[1, 3, 3, 1],
@@ -689,7 +689,7 @@ class PoolingTest(test.TestCase):
 
       for v2 in [True, False]:
         self._VerifyValues(
-            gen_nn_ops._max_pool_v2,
+            gen_nn_ops.max_pool_v2,
             input_sizes=[1, 3, 3, 1],
             ksize=[1, 1, 1, 1],
             strides=[1, 2, 2, 1],
@@ -699,7 +699,7 @@ class PoolingTest(test.TestCase):
             v2=v2)
 
         self._VerifyValues(
-            gen_nn_ops._max_pool_v2,
+            gen_nn_ops.max_pool_v2,
             input_sizes=[1, 4, 4, 1],
             ksize=[1, 1, 1, 1],
             strides=[1, 2, 2, 1],
@@ -764,8 +764,8 @@ class PoolingTest(test.TestCase):
         _, argmax_op = nn_ops.max_pool_with_argmax(t, ksize, strides, padding)
         argmax = argmax_op.eval()
         grad_in = constant_op.constant(tensor_output, shape=output_shape)
-        out_op = gen_nn_ops._max_pool_grad_with_argmax(t, grad_in, argmax,
-                                                       ksize, strides, padding)
+        out_op = gen_nn_ops.max_pool_grad_with_argmax(t, grad_in, argmax, ksize,
+                                                      strides, padding)
         gpu_val = out_op.eval()
         self.assertShapeEqual(gpu_val, out_op)
       with self.test_session(use_gpu=False):
@@ -773,8 +773,8 @@ class PoolingTest(test.TestCase):
         out_op = nn_ops.max_pool(t, ksize, strides, padding)
         orig_out = out_op.eval()
         grad_in = constant_op.constant(tensor_output, shape=output_shape)
-        out_op = gen_nn_ops._max_pool_grad(t, orig_out, grad_in, ksize, strides,
-                                           padding)
+        out_op = gen_nn_ops.max_pool_grad(t, orig_out, grad_in, ksize, strides,
+                                          padding)
         cpu_val = out_op.eval()
         self.assertShapeEqual(cpu_val, out_op)
       # The CPU version accumulates its gradient on fp16, so it's less
@@ -793,7 +793,7 @@ class PoolingTest(test.TestCase):
         _, argmax_op = nn_ops.max_pool_with_argmax(t, ksize, strides, padding)
         argmax = argmax_op.eval()
         grad_in = constant_op.constant(tensor_input, shape=input_shape)
-        out_op = gen_nn_ops._max_pool_grad_grad_with_argmax(
+        out_op = gen_nn_ops.max_pool_grad_grad_with_argmax(
             t, grad_in, argmax, ksize, strides, padding)
         gpu_val = out_op.eval()
         self.assertShapeEqual(gpu_val, out_op)
@@ -802,8 +802,8 @@ class PoolingTest(test.TestCase):
         out_op = nn_ops.max_pool(t, ksize, strides, padding)
         orig_out = out_op.eval()
         grad_in = constant_op.constant(tensor_input, shape=input_shape)
-        out_op = gen_nn_ops._max_pool_grad_grad(t, orig_out, grad_in, ksize,
-                                                strides, padding)
+        out_op = gen_nn_ops.max_pool_grad_grad(t, orig_out, grad_in, ksize,
+                                               strides, padding)
         cpu_val = out_op.eval()
         self.assertShapeEqual(cpu_val, out_op)
       # The CPU version accumulates its gradient on fp16, so it's less
@@ -842,7 +842,7 @@ class PoolingTest(test.TestCase):
       t = constant_op.constant(tensor_input, shape=[1, 2, 2, 1])
       argmax = constant_op.constant(
           tensor_argmax, shape=[1, 2, 2, 1], dtype=dtypes.int64)
-      out_op = gen_nn_ops._max_pool_grad_with_argmax(
+      out_op = gen_nn_ops.max_pool_grad_with_argmax(
           orig_in,
           t,
           argmax,
@@ -865,7 +865,7 @@ class PoolingTest(test.TestCase):
       t = constant_op.constant(tensor_input, shape=[1, 3, 3, 1])
       argmax = constant_op.constant(
           tensor_argmax, shape=[1, 2, 2, 1], dtype=dtypes.int64)
-      out_op = gen_nn_ops._max_pool_grad_grad_with_argmax(
+      out_op = gen_nn_ops.max_pool_grad_grad_with_argmax(
           orig_in,
           t,
           argmax,
@@ -1029,7 +1029,7 @@ class PoolingTest(test.TestCase):
     self.assertLess(err, err_tolerance)
 
   def _testMaxPoolGradValidPadding1_1(self, data_format, use_gpu):
-    for pool_func in [gen_nn_ops._max_pool_v2, nn_ops.max_pool]:
+    for pool_func in [gen_nn_ops.max_pool_v2, nn_ops.max_pool]:
       self._ConstructAndTestGradient(
           pool_func,
           input_sizes=[1, 3, 3, 1],
@@ -1043,7 +1043,7 @@ class PoolingTest(test.TestCase):
           use_gpu=use_gpu)
 
   def _testMaxPoolGradValidPadding2_1_6(self, data_format, use_gpu):
-    for pool_func in [gen_nn_ops._max_pool_v2, nn_ops.max_pool]:
+    for pool_func in [gen_nn_ops.max_pool_v2, nn_ops.max_pool]:
       self._ConstructAndTestGradient(
           pool_func,
           input_sizes=[2, 6, 6, 3],
@@ -1057,7 +1057,7 @@ class PoolingTest(test.TestCase):
           use_gpu=use_gpu)
 
   def _testMaxPoolGradValidPadding2_1_7(self, data_format, use_gpu):
-    for pool_func in [gen_nn_ops._max_pool_v2, nn_ops.max_pool]:
+    for pool_func in [gen_nn_ops.max_pool_v2, nn_ops.max_pool]:
       self._ConstructAndTestGradient(
           pool_func,
           input_sizes=[2, 7, 7, 3],
@@ -1071,7 +1071,7 @@ class PoolingTest(test.TestCase):
           use_gpu=use_gpu)
 
   def _testMaxPoolGradValidPadding1_2(self, data_format, use_gpu):
-    for pool_func in [gen_nn_ops._max_pool_v2, nn_ops.max_pool]:
+    for pool_func in [gen_nn_ops.max_pool_v2, nn_ops.max_pool]:
       self._ConstructAndTestGradient(
           pool_func,
           input_sizes=[1, 3, 3, 1],
@@ -1085,7 +1085,7 @@ class PoolingTest(test.TestCase):
           use_gpu=use_gpu)
 
   def _testMaxPoolGradValidPadding2_2(self, data_format, use_gpu):
-    for pool_func in [gen_nn_ops._max_pool_v2, nn_ops.max_pool]:
+    for pool_func in [gen_nn_ops.max_pool_v2, nn_ops.max_pool]:
       self._ConstructAndTestGradient(
           pool_func,
           input_sizes=[2, 2, 2, 3],
@@ -1099,7 +1099,7 @@ class PoolingTest(test.TestCase):
           use_gpu=use_gpu)
 
   def _testMaxPoolGradSamePadding1_1(self, data_format, use_gpu):
-    for pool_func in [gen_nn_ops._max_pool_v2, nn_ops.max_pool]:
+    for pool_func in [gen_nn_ops.max_pool_v2, nn_ops.max_pool]:
       self._ConstructAndTestGradient(
           pool_func,
           input_sizes=[2, 2, 4, 3],
@@ -1113,7 +1113,7 @@ class PoolingTest(test.TestCase):
           use_gpu=use_gpu)
 
   def _testMaxPoolGradSamePadding1_2(self, data_format, use_gpu):
-    for pool_func in [gen_nn_ops._max_pool_v2, nn_ops.max_pool]:
+    for pool_func in [gen_nn_ops.max_pool_v2, nn_ops.max_pool]:
       self._ConstructAndTestGradient(
           pool_func,
           input_sizes=[2, 2, 4, 3],
@@ -1127,7 +1127,7 @@ class PoolingTest(test.TestCase):
           use_gpu=use_gpu)
 
   def _testMaxPoolGradSamePadding2_1(self, data_format, use_gpu):
-    for pool_func in [gen_nn_ops._max_pool_v2, nn_ops.max_pool]:
+    for pool_func in [gen_nn_ops.max_pool_v2, nn_ops.max_pool]:
       self._ConstructAndTestGradient(
           pool_func,
           input_sizes=[2, 2, 4, 3],
@@ -1141,7 +1141,7 @@ class PoolingTest(test.TestCase):
           use_gpu=use_gpu)
 
   def _testMaxPoolGradSamePadding2_2(self, data_format, use_gpu):
-    for pool_func in [gen_nn_ops._max_pool_v2, nn_ops.max_pool]:
+    for pool_func in [gen_nn_ops.max_pool_v2, nn_ops.max_pool]:
       self._ConstructAndTestGradient(
           pool_func,
           input_sizes=[2, 2, 4, 3],
@@ -1155,7 +1155,7 @@ class PoolingTest(test.TestCase):
           use_gpu=use_gpu)
 
   def _testMaxPoolGradSamePadding3_1(self, data_format, use_gpu):
-    for pool_func in [gen_nn_ops._max_pool_v2, nn_ops.max_pool]:
+    for pool_func in [gen_nn_ops.max_pool_v2, nn_ops.max_pool]:
       self._ConstructAndTestGradient(
           pool_func,
           input_sizes=[1, 7, 7, 1],
@@ -1199,7 +1199,7 @@ class PoolingTest(test.TestCase):
     Returns:
       A Tensor.
     """
-    pool_func = gen_nn_ops.max_pool_grad_v2 if v2 else gen_nn_ops._max_pool_grad
+    pool_func = gen_nn_ops.max_pool_grad_v2 if v2 else gen_nn_ops.max_pool_grad
     return pool_func(orig_input, orig_output, grad,
                      [1, window_rows, window_cols, 1],
                      [1, row_stride, col_stride, 1], padding)
@@ -1208,7 +1208,7 @@ class PoolingTest(test.TestCase):
                              expected_input_backprop, input_sizes, output_sizes,
                              window_rows, window_cols, row_stride, col_stride,
                              padding, use_gpu, v2):
-    pool_func = gen_nn_ops._max_pool_v2 if v2 else nn_ops.max_pool
+    pool_func = gen_nn_ops.max_pool_v2 if v2 else nn_ops.max_pool
     with self.test_session(use_gpu=use_gpu):
       input_tensor = constant_op.constant(input_data, shape=input_sizes)
       output_tensor = pool_func(input_tensor, [1, window_rows, window_cols, 1],
@@ -1504,7 +1504,7 @@ class PoolingTest(test.TestCase):
     self._testMaxPoolGradDirectWithNans2_2()
 
   def _testMaxPoolGradGradValidPadding1_1(self, data_format, use_gpu):
-    for pool_func in [gen_nn_ops._max_pool_v2, nn_ops.max_pool]:
+    for pool_func in [gen_nn_ops.max_pool_v2, nn_ops.max_pool]:
       self._ConstructAndTestSecondGradient(
           pool_func,
           input_sizes=[1, 3, 3, 1],
@@ -1518,7 +1518,7 @@ class PoolingTest(test.TestCase):
           use_gpu=use_gpu)
 
   def _testMaxPoolGradGradValidPadding2_1_6(self, data_format, use_gpu):
-    for pool_func in [gen_nn_ops._max_pool_v2, nn_ops.max_pool]:
+    for pool_func in [gen_nn_ops.max_pool_v2, nn_ops.max_pool]:
       self._ConstructAndTestSecondGradient(
           pool_func,
           input_sizes=[2, 6, 6, 3],
@@ -1532,7 +1532,7 @@ class PoolingTest(test.TestCase):
           use_gpu=use_gpu)
 
   def _testMaxPoolGradGradValidPadding2_1_7(self, data_format, use_gpu):
-    for pool_func in [gen_nn_ops._max_pool_v2, nn_ops.max_pool]:
+    for pool_func in [gen_nn_ops.max_pool_v2, nn_ops.max_pool]:
       self._ConstructAndTestSecondGradient(
           pool_func,
           input_sizes=[2, 7, 7, 3],
@@ -1546,7 +1546,7 @@ class PoolingTest(test.TestCase):
           use_gpu=use_gpu)
 
   def _testMaxPoolGradGradValidPadding2_2(self, data_format, use_gpu):
-    for pool_func in [gen_nn_ops._max_pool_v2, nn_ops.max_pool]:
+    for pool_func in [gen_nn_ops.max_pool_v2, nn_ops.max_pool]:
       self._ConstructAndTestSecondGradient(
           pool_func,
           input_sizes=[2, 2, 2, 3],
@@ -1560,7 +1560,7 @@ class PoolingTest(test.TestCase):
           use_gpu=use_gpu)
 
   def _testMaxPoolGradGradSamePadding1_1(self, data_format, use_gpu):
-    for pool_func in [gen_nn_ops._max_pool_v2, nn_ops.max_pool]:
+    for pool_func in [gen_nn_ops.max_pool_v2, nn_ops.max_pool]:
       self._ConstructAndTestSecondGradient(
           pool_func,
           input_sizes=[2, 2, 4, 3],
@@ -1574,7 +1574,7 @@ class PoolingTest(test.TestCase):
           use_gpu=use_gpu)
 
   def _testMaxPoolGradGradSamePadding2_1(self, data_format, use_gpu):
-    for pool_func in [gen_nn_ops._max_pool_v2, nn_ops.max_pool]:
+    for pool_func in [gen_nn_ops.max_pool_v2, nn_ops.max_pool]:
       self._ConstructAndTestSecondGradient(
           pool_func,
           input_sizes=[2, 2, 4, 3],
@@ -1588,7 +1588,7 @@ class PoolingTest(test.TestCase):
           use_gpu=use_gpu)
 
   def _testMaxPoolGradGradSamePadding2_2(self, data_format, use_gpu):
-    for pool_func in [gen_nn_ops._max_pool_v2, nn_ops.max_pool]:
+    for pool_func in [gen_nn_ops.max_pool_v2, nn_ops.max_pool]:
       self._ConstructAndTestSecondGradient(
           pool_func,
           input_sizes=[2, 2, 4, 3],
@@ -1602,7 +1602,7 @@ class PoolingTest(test.TestCase):
           use_gpu=use_gpu)
 
   def _testMaxPoolGradGradSamePadding3_1(self, data_format, use_gpu):
-    for pool_func in [gen_nn_ops._max_pool_v2, nn_ops.max_pool]:
+    for pool_func in [gen_nn_ops.max_pool_v2, nn_ops.max_pool]:
       self._ConstructAndTestSecondGradient(
           pool_func,
           input_sizes=[1, 7, 7, 1],
@@ -1644,7 +1644,7 @@ class PoolingTest(test.TestCase):
     Returns:
       A Tensor.
     """
-    return gen_nn_ops._max_pool_grad_grad(
+    return gen_nn_ops.max_pool_grad_grad(
         orig_input, orig_output, grad, [1, window_rows, window_cols, 1],
         [1, row_stride, col_stride, 1], padding)
 
diff --git a/tensorflow/python/kernel_tests/save_restore_ops_test.py b/tensorflow/python/kernel_tests/save_restore_ops_test.py
index 1bdfa9ebd8..cb9aa1e34d 100644
--- a/tensorflow/python/kernel_tests/save_restore_ops_test.py
+++ b/tensorflow/python/kernel_tests/save_restore_ops_test.py
@@ -31,11 +31,10 @@ class ShardedFileOpsTest(test.TestCase):
     with session.Session(
         target="", config=config_pb2.ConfigProto(device_count={"CPU": 2})):
       self.assertEqual(
-          gen_io_ops._sharded_filename("foo", 4, 100).eval(),
+          gen_io_ops.sharded_filename("foo", 4, 100).eval(),
           b"foo-00004-of-00100")
       self.assertEqual(
-          gen_io_ops._sharded_filespec("foo", 100).eval(),
-          b"foo-?????-of-00100")
+          gen_io_ops.sharded_filespec("foo", 100).eval(), b"foo-?????-of-00100")
 
 
 class ShapeInferenceTest(test.TestCase):
@@ -53,7 +52,7 @@ class ShapeInferenceTest(test.TestCase):
                         [dtypes.float32, dtypes.float32])
 
   def testRestoreSlice(self):
-    op = gen_io_ops._restore_slice("model", "var", "3 4 0,1:-", dtypes.float32)
+    op = gen_io_ops.restore_slice("model", "var", "3 4 0,1:-", dtypes.float32)
     self.assertEqual([1, 4], op.get_shape())
 
 
diff --git a/tensorflow/python/kernel_tests/scalar_test.py b/tensorflow/python/kernel_tests/scalar_test.py
index e65241981e..0d8fd23294 100644
--- a/tensorflow/python/kernel_tests/scalar_test.py
+++ b/tensorflow/python/kernel_tests/scalar_test.py
@@ -92,11 +92,11 @@ class ScalarTest(test.TestCase):
     self.check(array_ops.reshape, (7, 1), 'sizes input must be 1-D', [7])
 
   def testShardedFilename(self):
-    self.check(gen_io_ops._sharded_filename, ('foo', 4, [100]),
+    self.check(gen_io_ops.sharded_filename, ('foo', 4, [100]),
                'must be a scalar', b'foo-00004-of-00100')
 
   def testShardedFilespec(self):
-    self.check(gen_io_ops._sharded_filespec, ('foo', [100]), 'must be a scalar',
+    self.check(gen_io_ops.sharded_filespec, ('foo', [100]), 'must be a scalar',
                b'foo-?????-of-00100')
 
   def testUnsortedSegmentSum(self):
diff --git a/tensorflow/python/kernel_tests/spacetobatch_op_test.py b/tensorflow/python/kernel_tests/spacetobatch_op_test.py
index b943dfa4e5..2a9232b6ae 100644
--- a/tensorflow/python/kernel_tests/spacetobatch_op_test.py
+++ b/tensorflow/python/kernel_tests/spacetobatch_op_test.py
@@ -86,11 +86,11 @@ class CppOpImpl(object):
 
   @staticmethod
   def space_to_batch(*args, **kwargs):
-    return gen_array_ops._space_to_batch(*args, **kwargs)
+    return gen_array_ops.space_to_batch(*args, **kwargs)
 
   @staticmethod
   def batch_to_space(*args, **kwargs):
-    return gen_array_ops._batch_to_space(*args, **kwargs)
+    return gen_array_ops.batch_to_space(*args, **kwargs)
 
 
 class SpaceToBatchTest(test.TestCase, PythonOpImpl):
diff --git a/tensorflow/python/kernel_tests/sparse_xent_op_test.py b/tensorflow/python/kernel_tests/sparse_xent_op_test.py
index cd5b711a0e..a841fe83a7 100644
--- a/tensorflow/python/kernel_tests/sparse_xent_op_test.py
+++ b/tensorflow/python/kernel_tests/sparse_xent_op_test.py
@@ -64,7 +64,7 @@ class SparseXentTest(test.TestCase):
   def _testXent(self, np_features, np_labels):
     np_loss, np_backprop = self._npXent(np_features, np_labels)
     with self.test_session(use_gpu=True) as sess:
-      loss, backprop = gen_nn_ops._sparse_softmax_cross_entropy_with_logits(
+      loss, backprop = gen_nn_ops.sparse_softmax_cross_entropy_with_logits(
           np_features, np_labels)
       tf_loss, tf_backprop = sess.run([loss, backprop])
     self.assertAllCloseAccordingToType(np_loss, tf_loss)
@@ -73,7 +73,7 @@ class SparseXentTest(test.TestCase):
   def testSingleClass(self):
     for label_dtype in np.int32, np.int64:
       with self.test_session(use_gpu=True) as sess:
-        loss, backprop = gen_nn_ops._sparse_softmax_cross_entropy_with_logits(
+        loss, backprop = gen_nn_ops.sparse_softmax_cross_entropy_with_logits(
             np.array([[1.], [-1.], [0.]]).astype(np.float32),
             np.array([0, 0, 0]).astype(label_dtype))
         tf_loss, tf_backprop = sess.run([loss, backprop])
@@ -87,8 +87,9 @@ class SparseXentTest(test.TestCase):
 
     if test.is_built_with_cuda() and test.is_gpu_available():
       with self.test_session(use_gpu=True) as sess:
-        loss, backprop = (gen_nn_ops._sparse_softmax_cross_entropy_with_logits(
-            features, labels))
+        loss, backprop = (
+            gen_nn_ops.sparse_softmax_cross_entropy_with_logits(
+                features, labels))
         tf_loss, tf_backprop = sess.run([loss, backprop])
         self.assertAllClose(
             [[np.nan] * 4, [0.25, 0.25, 0.25, -0.75],
@@ -100,8 +101,8 @@ class SparseXentTest(test.TestCase):
             [np.nan, 1.3862, 3.4420, np.nan], tf_loss, rtol=1e-3, atol=1e-3)
 
     with self.test_session(use_gpu=False) as sess:
-      loss, backprop = (gen_nn_ops._sparse_softmax_cross_entropy_with_logits(
-          features, labels))
+      loss, backprop = (
+          gen_nn_ops.sparse_softmax_cross_entropy_with_logits(features, labels))
       with self.assertRaisesOpError("Received a label value of"):
         sess.run([loss, backprop])
 
diff --git a/tensorflow/python/kernel_tests/stack_ops_test.py b/tensorflow/python/kernel_tests/stack_ops_test.py
index aa409336f5..afd2eaffab 100644
--- a/tensorflow/python/kernel_tests/stack_ops_test.py
+++ b/tensorflow/python/kernel_tests/stack_ops_test.py
@@ -34,11 +34,11 @@ class StackOpTest(test.TestCase):
 
   def _testStackPushPop(self, use_gpu):
     with self.test_session(use_gpu=use_gpu):
-      h = gen_data_flow_ops._stack_v2(
+      h = gen_data_flow_ops.stack_v2(
           -1, elem_type=dtypes.float32, stack_name="foo")
-      c = gen_data_flow_ops._stack_push_v2(h, [[4.0, 5.0]])
+      c = gen_data_flow_ops.stack_push_v2(h, [[4.0, 5.0]])
       with ops.control_dependencies([c]):
-        c1 = gen_data_flow_ops._stack_pop_v2(h, dtypes.float32)
+        c1 = gen_data_flow_ops.stack_pop_v2(h, dtypes.float32)
       self.assertAllClose([[4.0, 5.0]], c1.eval())
 
   def testStackPushPop(self):
@@ -49,11 +49,11 @@ class StackOpTest(test.TestCase):
     with self.test_session(use_gpu=use_gpu):
       a = np.arange(2000)
       x = constant_op.constant(a, dtype=dtypes.float32)
-      h = gen_data_flow_ops._stack_v2(
+      h = gen_data_flow_ops.stack_v2(
           -1, elem_type=dtypes.float32, stack_name="foo")
-      c = gen_data_flow_ops._stack_push_v2(h, x, swap_memory=True)
+      c = gen_data_flow_ops.stack_push_v2(h, x, swap_memory=True)
       with ops.control_dependencies([c]):
-        c1 = gen_data_flow_ops._stack_pop_v2(h, dtypes.float32)
+        c1 = gen_data_flow_ops.stack_pop_v2(h, dtypes.float32)
       self.assertAllClose(a, c1.eval())
 
   def testStackPushPopSwap(self):
@@ -63,7 +63,7 @@ class StackOpTest(test.TestCase):
   def _testStackWhileSwap(self, use_gpu):
     with self.test_session(use_gpu=use_gpu):
       n = constant_op.constant(0)
-      h = gen_data_flow_ops._stack_v2(
+      h = gen_data_flow_ops.stack_v2(
           -1, elem_type=dtypes.float32, stack_name="foo")
 
       def c(x):
@@ -72,7 +72,7 @@ class StackOpTest(test.TestCase):
       def b(x):
         with ops.control_dependencies([x]):
           a = constant_op.constant(np.ones(2000), dtype=dtypes.float32)
-          v = gen_data_flow_ops._stack_push_v2(h, a, swap_memory=True)
+          v = gen_data_flow_ops.stack_push_v2(h, a, swap_memory=True)
         with ops.control_dependencies([v]):
           return math_ops.add(x, 1)
 
@@ -86,7 +86,7 @@ class StackOpTest(test.TestCase):
 
       def b1(x, y):
         nx = math_ops.subtract(x, 1)
-        ny = y + gen_data_flow_ops._stack_pop_v2(h, dtypes.float32)
+        ny = y + gen_data_flow_ops.stack_pop_v2(h, dtypes.float32)
         return [nx, ny]
 
       _, ry = control_flow_ops.while_loop(
@@ -99,16 +99,16 @@ class StackOpTest(test.TestCase):
 
   def _testMultiStack(self, use_gpu):
     with self.test_session(use_gpu=use_gpu):
-      h1 = gen_data_flow_ops._stack_v2(
+      h1 = gen_data_flow_ops.stack_v2(
           -1, elem_type=dtypes.float32, stack_name="foo")
-      c1 = gen_data_flow_ops._stack_push_v2(h1, 4.0)
+      c1 = gen_data_flow_ops.stack_push_v2(h1, 4.0)
       with ops.control_dependencies([c1]):
-        c1 = gen_data_flow_ops._stack_pop_v2(h1, dtypes.float32)
-      h2 = gen_data_flow_ops._stack_v2(
+        c1 = gen_data_flow_ops.stack_pop_v2(h1, dtypes.float32)
+      h2 = gen_data_flow_ops.stack_v2(
           -1, elem_type=dtypes.float32, stack_name="bar")
-      c2 = gen_data_flow_ops._stack_push_v2(h2, 5.0)
+      c2 = gen_data_flow_ops.stack_push_v2(h2, 5.0)
       with ops.control_dependencies([c2]):
-        c2 = gen_data_flow_ops._stack_pop_v2(h2, dtypes.float32)
+        c2 = gen_data_flow_ops.stack_pop_v2(h2, dtypes.float32)
       r = c1 + c2
       self.assertAllClose(9.0, r.eval())
 
@@ -119,17 +119,17 @@ class StackOpTest(test.TestCase):
   def _testSameNameStacks(self, use_gpu):
     """Different stacks with the same name do not interfere."""
     with self.test_session(use_gpu=use_gpu) as sess:
-      h1 = gen_data_flow_ops._stack_v2(
+      h1 = gen_data_flow_ops.stack_v2(
           -1, elem_type=dtypes.float32, stack_name="foo")
-      h2 = gen_data_flow_ops._stack_v2(
+      h2 = gen_data_flow_ops.stack_v2(
           -1, elem_type=dtypes.float32, stack_name="foo")
 
-      c1 = gen_data_flow_ops._stack_push_v2(h1, 4.0)
+      c1 = gen_data_flow_ops.stack_push_v2(h1, 4.0)
       with ops.control_dependencies([c1]):
-        c2 = gen_data_flow_ops._stack_push_v2(h2, 5.0)
+        c2 = gen_data_flow_ops.stack_push_v2(h2, 5.0)
       with ops.control_dependencies([c2]):
-        pop1 = gen_data_flow_ops._stack_pop_v2(h1, dtypes.float32)
-        pop2 = gen_data_flow_ops._stack_pop_v2(h2, dtypes.float32)
+        pop1 = gen_data_flow_ops.stack_pop_v2(h1, dtypes.float32)
+        pop2 = gen_data_flow_ops.stack_pop_v2(h2, dtypes.float32)
 
       out1, out2 = sess.run([pop1, pop2])
       self.assertAllClose(out1, 4.0)
@@ -141,9 +141,9 @@ class StackOpTest(test.TestCase):
 
   def _testCloseStack(self, use_gpu):
     with self.test_session(use_gpu=use_gpu) as sess:
-      h = gen_data_flow_ops._stack_v2(
+      h = gen_data_flow_ops.stack_v2(
           -1, elem_type=dtypes.float32, stack_name="foo")
-      c1 = gen_data_flow_ops._stack_close_v2(h)
+      c1 = gen_data_flow_ops.stack_close_v2(h)
       sess.run(c1)
 
   def testCloseStack(self):
@@ -152,11 +152,11 @@ class StackOpTest(test.TestCase):
 
   def _testPushCloseStack(self, use_gpu):
     with self.test_session(use_gpu=use_gpu) as sess:
-      h = gen_data_flow_ops._stack_v2(
+      h = gen_data_flow_ops.stack_v2(
           -1, elem_type=dtypes.float32, stack_name="foo")
-      c = gen_data_flow_ops._stack_push_v2(h, [[4.0, 5.0]])
+      c = gen_data_flow_ops.stack_push_v2(h, [[4.0, 5.0]])
       with ops.control_dependencies([c]):
-        c1 = gen_data_flow_ops._stack_close_v2(h)
+        c1 = gen_data_flow_ops.stack_close_v2(h)
       sess.run(c1)
 
   def testPushCloseStack(self):
@@ -170,9 +170,9 @@ class StackOpRefTest(test.TestCase):
   def _testStackPushPop(self, use_gpu):
     with self.test_session(use_gpu=use_gpu):
       h = gen_data_flow_ops._stack(dtypes.float32, stack_name="foo")
-      c = gen_data_flow_ops._stack_push(h, [[4.0, 5.0]])
+      c = gen_data_flow_ops.stack_push(h, [[4.0, 5.0]])
       with ops.control_dependencies([c]):
-        c1 = gen_data_flow_ops._stack_pop(h, dtypes.float32)
+        c1 = gen_data_flow_ops.stack_pop(h, dtypes.float32)
       self.assertAllClose([[4.0, 5.0]], c1.eval())
 
   def testStackPushPop(self):
@@ -184,9 +184,9 @@ class StackOpRefTest(test.TestCase):
       a = np.arange(2000)
       x = constant_op.constant(a, dtype=dtypes.float32)
       h = gen_data_flow_ops._stack(dtypes.float32, stack_name="foo")
-      c = gen_data_flow_ops._stack_push(h, x, swap_memory=True)
+      c = gen_data_flow_ops.stack_push(h, x, swap_memory=True)
       with ops.control_dependencies([c]):
-        c1 = gen_data_flow_ops._stack_pop(h, dtypes.float32)
+        c1 = gen_data_flow_ops.stack_pop(h, dtypes.float32)
       self.assertAllClose(a, c1.eval())
 
   def testStackPushPopSwap(self):
@@ -196,13 +196,13 @@ class StackOpRefTest(test.TestCase):
   def _testMultiStack(self, use_gpu):
     with self.test_session(use_gpu=use_gpu):
       h1 = gen_data_flow_ops._stack(dtypes.float32, stack_name="foo")
-      c1 = gen_data_flow_ops._stack_push(h1, 4.0)
+      c1 = gen_data_flow_ops.stack_push(h1, 4.0)
       with ops.control_dependencies([c1]):
-        c1 = gen_data_flow_ops._stack_pop(h1, dtypes.float32)
+        c1 = gen_data_flow_ops.stack_pop(h1, dtypes.float32)
       h2 = gen_data_flow_ops._stack(dtypes.float32, stack_name="bar")
-      c2 = gen_data_flow_ops._stack_push(h2, 5.0)
+      c2 = gen_data_flow_ops.stack_push(h2, 5.0)
       with ops.control_dependencies([c2]):
-        c2 = gen_data_flow_ops._stack_pop(h2, dtypes.float32)
+        c2 = gen_data_flow_ops.stack_pop(h2, dtypes.float32)
       r = c1 + c2
       self.assertAllClose(9.0, r.eval())
 
@@ -217,7 +217,7 @@ class StackOpRefTest(test.TestCase):
       def b(x):
         with ops.control_dependencies([x]):
           a = constant_op.constant(np.ones(2000), dtype=dtypes.float32)
-          v = gen_data_flow_ops._stack_push(h, a, swap_memory=True)
+          v = gen_data_flow_ops.stack_push(h, a, swap_memory=True)
         with ops.control_dependencies([v]):
           return math_ops.add(x, 1)
 
@@ -231,7 +231,7 @@ class StackOpRefTest(test.TestCase):
 
       def b1(x, y):
         nx = math_ops.subtract(x, 1)
-        ny = y + gen_data_flow_ops._stack_pop(h, dtypes.float32)
+        ny = y + gen_data_flow_ops.stack_pop(h, dtypes.float32)
         return [nx, ny]
 
       _, ry = control_flow_ops.while_loop(
@@ -249,9 +249,9 @@ class StackOpRefTest(test.TestCase):
   def _testSameNameStacks(self, use_gpu):
     with self.test_session(use_gpu=use_gpu):
       h1 = gen_data_flow_ops._stack(dtypes.float32, stack_name="foo")
-      c1 = gen_data_flow_ops._stack_push(h1, 4.0)
+      c1 = gen_data_flow_ops.stack_push(h1, 4.0)
       h2 = gen_data_flow_ops._stack(dtypes.float32, stack_name="foo")
-      c2 = gen_data_flow_ops._stack_push(h2, 5.0)
+      c2 = gen_data_flow_ops.stack_push(h2, 5.0)
       _ = c1 + c2
       self.assertNotEqual(h1.eval()[1], h2.eval()[1])
 
@@ -262,7 +262,7 @@ class StackOpRefTest(test.TestCase):
   def _testCloseStack(self, use_gpu):
     with self.test_session(use_gpu=use_gpu) as sess:
       h = gen_data_flow_ops._stack(dtypes.float32, stack_name="foo")
-      c1 = gen_data_flow_ops._stack_close(h)
+      c1 = gen_data_flow_ops.stack_close(h)
       sess.run(c1)
 
   def testCloseStack(self):
@@ -272,9 +272,9 @@ class StackOpRefTest(test.TestCase):
   def _testPushCloseStack(self, use_gpu):
     with self.test_session(use_gpu=use_gpu) as sess:
       h = gen_data_flow_ops._stack(dtypes.float32, stack_name="foo")
-      c = gen_data_flow_ops._stack_push(h, [[4.0, 5.0]])
+      c = gen_data_flow_ops.stack_push(h, [[4.0, 5.0]])
       with ops.control_dependencies([c]):
-        c1 = gen_data_flow_ops._stack_close(h)
+        c1 = gen_data_flow_ops.stack_close(h)
       sess.run(c1)
 
   def testPushCloseStack(self):
diff --git a/tensorflow/python/kernel_tests/tensor_array_ops_test.py b/tensorflow/python/kernel_tests/tensor_array_ops_test.py
index aad2443eea..8f09f3d78b 100644
--- a/tensorflow/python/kernel_tests/tensor_array_ops_test.py
+++ b/tensorflow/python/kernel_tests/tensor_array_ops_test.py
@@ -437,7 +437,7 @@ class TensorArrayTest(test.TestCase):
 
       # Test reading wrong datatype, which is only possible in graph mode
       if context.in_graph_mode():
-        r0_bad = gen_data_flow_ops._tensor_array_read_v3(
+        r0_bad = gen_data_flow_ops.tensor_array_read_v3(
             handle=w0.handle, index=0, dtype=dtypes.float64, flow_in=w0.flow)
         with self.assertRaisesOpError(
             "TensorArray dtype is float but Op requested dtype double."):
diff --git a/tensorflow/python/kernel_tests/unique_op_test.py b/tensorflow/python/kernel_tests/unique_op_test.py
index 6366d2e181..173d95b258 100644
--- a/tensorflow/python/kernel_tests/unique_op_test.py
+++ b/tensorflow/python/kernel_tests/unique_op_test.py
@@ -66,9 +66,9 @@ class UniqueTest(test.TestCase):
     for dtype in [np.int32, np.int64]:
       x = np.array([[1, 0, 0], [1, 0, 0], [2, 0, 0]])
       with self.test_session() as sess:
-        y0, idx0 = gen_array_ops._unique_v2(x, axis=np.array([0], dtype))
+        y0, idx0 = gen_array_ops.unique_v2(x, axis=np.array([0], dtype))
         tf_y0, tf_idx0 = sess.run([y0, idx0])
-        y1, idx1 = gen_array_ops._unique_v2(x, axis=np.array([1], dtype))
+        y1, idx1 = gen_array_ops.unique_v2(x, axis=np.array([1], dtype))
         tf_y1, tf_idx1 = sess.run([y1, idx1])
       self.assertAllEqual(tf_y0, np.array([[1, 0, 0], [2, 0, 0]]))
       self.assertAllEqual(tf_idx0, np.array([0, 0, 1]))
@@ -80,7 +80,7 @@ class UniqueTest(test.TestCase):
     # by default, the axis will be wrapped to allow `axis=None`.
     x = np.random.randint(2, high=10, size=7000)
     with self.test_session() as sess:
-      y, idx = gen_array_ops._unique_v2(x, axis=np.array([], np.int32))
+      y, idx = gen_array_ops.unique_v2(x, axis=np.array([], np.int32))
       tf_y, tf_idx = sess.run([y, idx])
 
     self.assertEqual(len(x), len(tf_idx))
diff --git a/tensorflow/python/kernel_tests/variable_ops_test.py b/tensorflow/python/kernel_tests/variable_ops_test.py
index 79071029fd..cf369c0718 100644
--- a/tensorflow/python/kernel_tests/variable_ops_test.py
+++ b/tensorflow/python/kernel_tests/variable_ops_test.py
@@ -165,26 +165,26 @@ class VariableOpTest(test.TestCase):
 
   def testTemporaryVariable(self):
     with self.test_session(use_gpu=True):
-      var = gen_state_ops._temporary_variable(
+      var = gen_state_ops.temporary_variable(
           [1, 2], dtypes.float32, var_name="foo")
       var = state_ops.assign(var, [[4.0, 5.0]])
       var = state_ops.assign_add(var, [[6.0, 7.0]])
-      final = gen_state_ops._destroy_temporary_variable(var, var_name="foo")
+      final = gen_state_ops.destroy_temporary_variable(var, var_name="foo")
       self.assertAllClose([[10.0, 12.0]], final.eval())
 
   def testDestroyNonexistentTemporaryVariable(self):
     with self.test_session(use_gpu=True):
-      var = gen_state_ops._temporary_variable([1, 2], dtypes.float32)
-      final = gen_state_ops._destroy_temporary_variable(var, var_name="bad")
+      var = gen_state_ops.temporary_variable([1, 2], dtypes.float32)
+      final = gen_state_ops.destroy_temporary_variable(var, var_name="bad")
       with self.assertRaises(errors.NotFoundError):
         final.eval()
 
   def testDuplicateTemporaryVariable(self):
     with self.test_session(use_gpu=True):
-      var1 = gen_state_ops._temporary_variable(
+      var1 = gen_state_ops.temporary_variable(
           [1, 2], dtypes.float32, var_name="dup")
       var1 = state_ops.assign(var1, [[1.0, 2.0]])
-      var2 = gen_state_ops._temporary_variable(
+      var2 = gen_state_ops.temporary_variable(
           [1, 2], dtypes.float32, var_name="dup")
       var2 = state_ops.assign(var2, [[3.0, 4.0]])
       final = var1 + var2
@@ -193,25 +193,25 @@ class VariableOpTest(test.TestCase):
 
   def testDestroyTemporaryVariableTwice(self):
     with self.test_session(use_gpu=True):
-      var = gen_state_ops._temporary_variable([1, 2], dtypes.float32)
-      val1 = gen_state_ops._destroy_temporary_variable(var, var_name="dup")
-      val2 = gen_state_ops._destroy_temporary_variable(var, var_name="dup")
+      var = gen_state_ops.temporary_variable([1, 2], dtypes.float32)
+      val1 = gen_state_ops.destroy_temporary_variable(var, var_name="dup")
+      val2 = gen_state_ops.destroy_temporary_variable(var, var_name="dup")
       final = val1 + val2
       with self.assertRaises(errors.NotFoundError):
         final.eval()
 
   def testTemporaryVariableNoLeak(self):
     with self.test_session(use_gpu=True):
-      var = gen_state_ops._temporary_variable(
+      var = gen_state_ops.temporary_variable(
           [1, 2], dtypes.float32, var_name="bar")
       final = array_ops.identity(var)
       final.eval()
 
   def testTwoTemporaryVariablesNoLeaks(self):
     with self.test_session(use_gpu=True):
-      var1 = gen_state_ops._temporary_variable(
+      var1 = gen_state_ops.temporary_variable(
           [1, 2], dtypes.float32, var_name="var1")
-      var2 = gen_state_ops._temporary_variable(
+      var2 = gen_state_ops.temporary_variable(
           [1, 2], dtypes.float32, var_name="var2")
       final = var1 + var2
       final.eval()
diff --git a/tensorflow/python/kernel_tests/variables_test.py b/tensorflow/python/kernel_tests/variables_test.py
index b16c8c002c..27599868b7 100644
--- a/tensorflow/python/kernel_tests/variables_test.py
+++ b/tensorflow/python/kernel_tests/variables_test.py
@@ -687,7 +687,7 @@ class VariableContainerTest(test.TestCase):
         v1 = variables.Variable([1])
         with ops.container("l2"):
           v2 = variables.Variable([2])
-          special_v = gen_state_ops._variable(
+          special_v = gen_state_ops.variable(
               shape=[1],
               dtype=dtypes.float32,
               name="VariableInL3",
diff --git a/tensorflow/python/kernel_tests/xent_op_test.py b/tensorflow/python/kernel_tests/xent_op_test.py
index e152f02d8e..e3e120a4eb 100644
--- a/tensorflow/python/kernel_tests/xent_op_test.py
+++ b/tensorflow/python/kernel_tests/xent_op_test.py
@@ -48,7 +48,7 @@ class XentTest(test.TestCase):
   def _testXent(self, np_features, np_labels, use_gpu=False):
     np_loss, np_backprop = self._npXent(np_features, np_labels)
     with self.test_session(use_gpu=use_gpu) as sess:
-      loss, backprop = gen_nn_ops._softmax_cross_entropy_with_logits(
+      loss, backprop = gen_nn_ops.softmax_cross_entropy_with_logits(
           np_features, np_labels)
       tf_loss, tf_backprop = sess.run([loss, backprop])
     self.assertAllCloseAccordingToType(np_loss, tf_loss)
@@ -71,7 +71,7 @@ class XentTest(test.TestCase):
   def _testSingleClass(self, use_gpu=False):
     for dtype in np.float16, np.float32:
       with self.test_session(use_gpu=use_gpu) as sess:
-        loss, backprop = gen_nn_ops._softmax_cross_entropy_with_logits(
+        loss, backprop = gen_nn_ops.softmax_cross_entropy_with_logits(
             np.array([[1.], [-1.], [0.]]).astype(dtype),
             np.array([[-1.], [0.], [1.]]).astype(dtype))
         tf_loss, tf_backprop = sess.run([loss, backprop])
@@ -89,7 +89,7 @@ class XentTest(test.TestCase):
       np_labels = np.array([[[0., 0., 0., 1.]], [[0., .5, .5,
                                                   0.]]]).astype(dtype)
       self.assertRaisesRegexp(ValueError, "must be rank 2",
-                              gen_nn_ops._softmax_cross_entropy_with_logits,
+                              gen_nn_ops.softmax_cross_entropy_with_logits,
                               np_features, np_labels)
 
   def testNpXent(self):
@@ -131,14 +131,14 @@ class XentTest(test.TestCase):
   def testShapeMismatch(self):
     with self.test_session():
       with self.assertRaises(ValueError):
-        gen_nn_ops._softmax_cross_entropy_with_logits(
+        gen_nn_ops.softmax_cross_entropy_with_logits(
             [[0., 1.], [2., 3.]], [[0., 1., 0.], [1., 0., 0.]])
 
   def testNotMatrix(self):
     with self.test_session():
       with self.assertRaises(ValueError):
-        gen_nn_ops._softmax_cross_entropy_with_logits([0., 1., 2., 3.],
-                                                      [0., 1., 0., 1.])
+        gen_nn_ops.softmax_cross_entropy_with_logits([0., 1., 2., 3.],
+                                                     [0., 1., 0., 1.])
 
   def testHalf(self):
     self._testAll(