Method test3DInputAxis2

tensorflow/python/layers/normalization_test.py:361–400

(self)

      self.assertAlmostEqual(np.std(normed_np_output), 1., places=1)

  def test3DInputAxis2(self):
    epsilon = 1e-3
    bn = normalization_layers.BatchNormalization(
        axis=2, epsilon=epsilon, momentum=0.9)
    inputs = variables.Variable(
        np.random.random((5, 4, 3)) + 100, dtype=dtypes.float32)
    training = array_ops.placeholder(dtype='bool')
    outputs = bn.apply(inputs, training=training)

    with self.cached_session() as sess:
      # Test training with placeholder learning phase.
      self.evaluate(variables.global_variables_initializer())
      np_gamma, np_beta = self.evaluate([bn.gamma, bn.beta])
      np_gamma = np.reshape(np_gamma, (1, 1, 3))
      np_beta = np.reshape(np_beta, (1, 1, 3))
      for _ in range(100):
        np_output, _, _ = sess.run([outputs] + bn.updates,
                                   feed_dict={training: True})
        # Verify that the axis is normalized during training.
        normed_np_output = ((np_output - epsilon) * np_gamma) + np_beta
        self.assertAlmostEqual(np.mean(normed_np_output), 0., places=1)
        self.assertAlmostEqual(np.std(normed_np_output), 1., places=1)

      # Verify that the statistics are updated during training.
      moving_mean, moving_var = self.evaluate(
          [bn.moving_mean, bn.moving_variance])
      np_inputs = self.evaluate(inputs)
      mean = np.mean(np_inputs, axis=(0, 1))
      std = np.std(np_inputs, axis=(0, 1))
      variance = np.square(std)
      self.assertAllClose(mean, moving_mean, atol=1e-2)
      self.assertAllClose(variance, moving_var, atol=1e-2)

      # Test inference with placeholder learning phase.
      np_output = sess.run(outputs, feed_dict={training: False})

      # Verify that the axis is normalized during inference.
      normed_np_output = ((np_output - epsilon) * np_gamma) + np_beta
      self.assertAlmostEqual(np.mean(normed_np_output), 0., places=1)
      self.assertAlmostEqual(np.std(normed_np_output), 1., places=1)

  def test4DInputAxis1(self):
    if test.is_gpu_available(cuda_only=True):