Fix L2Normalize when passing a list of dims. Fixes tensorflow#3932.

Change: 130947456

Author: Vijay Vasudevan

tensorflow/python/ops/nn.py

@@ -547,7 +547,8 @@ def l2_normalize(x, dim, epsilon=1e-12, name=None):
 
   Args:
     x: A `Tensor`.
-    dim: Dimension along which to normalize.
+    dim: Dimension along which to normalize.  A scalar or a vector of
+      integers.
     epsilon: A lower bound value for the norm. Will use `sqrt(epsilon)` as the
       divisor if `norm < sqrt(epsilon)`.
     name: A name for this operation (optional).
@@ -557,7 +558,7 @@ def l2_normalize(x, dim, epsilon=1e-12, name=None):
   """
   with ops.name_scope(name, "l2_normalize", [x]) as name:
     x = ops.convert_to_tensor(x, name="x")
-    square_sum = math_ops.reduce_sum(math_ops.square(x), [dim], keep_dims=True)
+    square_sum = math_ops.reduce_sum(math_ops.square(x), dim, keep_dims=True)
     x_inv_norm = math_ops.rsqrt(math_ops.maximum(square_sum, epsilon))
     return math_ops.mul(x, x_inv_norm, name=name)
 

tensorflow/python/ops/nn_test.py

@@ -180,8 +180,14 @@ def testGradient(self):
 class L2NormalizeTest(tf.test.TestCase):
 
   def _l2Normalize(self, x, dim):
-    norm = np.apply_along_axis(np.linalg.norm, dim, x)
-    return x / np.expand_dims(norm, dim)
+    if isinstance(dim, list):
+      norm = np.linalg.norm(x, axis=tuple(dim))
+      for d in dim:
+        norm = np.expand_dims(norm, d)
+      return x / norm
+    else:
+      norm = np.apply_along_axis(np.linalg.norm, dim, x)
+      return x / np.expand_dims(norm, dim)
 
   def testL2Normalize(self):
     x_shape = [20, 7, 3]
@@ -194,6 +200,17 @@ def testL2Normalize(self):
         y_tf = tf.nn.l2_normalize(x_tf, dim)
         self.assertAllClose(y_np, y_tf.eval())
 
+  def testL2NormalizeDimArray(self):
+    x_shape = [20, 7, 3]
+    np.random.seed(1)
+    x_np = np.random.random_sample(x_shape).astype(np.float32)
+    dim = [1, 2]
+    y_np = self._l2Normalize(x_np, dim)
+    with self.test_session():
+      x_tf = tf.constant(x_np, name="x")
+      y_tf = tf.nn.l2_normalize(x_tf, dim)
+      self.assertAllClose(y_np, y_tf.eval())
+
   def testL2NormalizeGradient(self):
     x_shape = [20, 7, 3]
     np.random.seed(1)