def test_variance_scaling_initializer_dtypes(self):
initializer0 = depthwise_initializers.DepthwiseVarianceScaling()
tensor0 = initializer0([3, 3, 128, 1])
self.assertEqual(tensor0.dtype, tf.float32)
initializer1 = depthwise_initializers.DepthwiseVarianceScaling()
tensor1 = initializer1([3, 3, 128, 1], dtype=tf.float64)
self.assertEqual(tensor1.dtype, tf.float64)
initializer2 = depthwise_initializers.DepthwiseVarianceScaling(
dtype=tf.float64)
tensor2 = initializer2([3, 3, 128, 1])
self.assertEqual(tensor2.dtype, tf.float64)
def test_depthwise_variance_scaling_end_to_end(self):
# This is an end-to-end test for the VarianceScaling() class.
# We apply he initializer to a tensor, and verify that the
# distribution of outputs matches what we expect.
input_tensor = tf.random.normal(
shape=(32, 20, 20, 1024),
mean=0.0,
stddev=1)
kernel_initializer = depthwise_initializers.DepthwiseVarianceScaling(
scale=1.0,
mode='fan_in',
distribution='truncated_normal')
kernel = tf.get_variable(
name='kernel',
initializer=kernel_initializer,
shape=[5, 5, 1024, 1])
output_tensor = tf.nn.depthwise_conv2d(
input_tensor,
kernel,
strides=(1, 1, 1, 1),
padding='VALID')
self.evaluate(tf.global_variables_initializer())
result = self.evaluate(output_tensor)
self.assertNear(np.mean(result), 0.0, 0.05)
self.assertNear(np.std(result), 1.0, 0.05)
def test_fan_avg_depth_multiplier_is_2(self):
initializer = depthwise_initializers.DepthwiseVarianceScaling(
scale=1.0, mode='fan_avg', distribution='untruncated_normal')
tensor = initializer([3, 5, 1024, 2])
value = self.evaluate(tensor)
self.assertEqual(value.shape, (3, 5, 1024, 2))
self.assertNear(np.mean(value), 0.0, 0.01)
self.assertNear(np.std(value), 1.0 / math.sqrt(1.5 * 3 * 5), 0.01)
def test_variance_scaling_scale_is_2(self):
initializer = depthwise_initializers.DepthwiseVarianceScaling(
scale=2.0, mode='fan_in', distribution='untruncated_normal')
tensor = initializer([3, 5, 1024, 1])
value = self.evaluate(tensor)
self.assertEqual(value.shape, (3, 5, 1024, 1))
self.assertNear(np.mean(value), 0.0, 0.01)
self.assertNear(np.std(value), math.sqrt(2.0 / (3 * 5)), 0.01)
def test_variance_scaling_uniform_fan_in(self):
initializer = depthwise_initializers.DepthwiseVarianceScaling(
scale=1.0, mode='fan_in', distribution='uniform')
tensor = initializer([3, 5, 1024, 1])
value = self.evaluate(tensor)
self.assertEqual(value.shape, (3, 5, 1024, 1))
self.assertNear(np.mean(value), 0.0, 0.01)
self.assertNear(np.std(value), 1.0 / math.sqrt(3 * 5), 0.01)
def test_variance_scaling_seed(self):
initializer = depthwise_initializers.DepthwiseVarianceScaling(seed=42)
tensor1 = initializer([3, 3, 128, 1])
tensor2 = initializer([3, 3, 128, 1])
self.assertAllClose(self.evaluate(tensor1), self.evaluate(tensor2))
