Python tensorflow.python.ops.gen_nn_ops._batch_norm_with_global_normalization_grad() Examples
The following are 6
code examples of tensorflow.python.ops.gen_nn_ops._batch_norm_with_global_normalization_grad().
You can vote up the ones you like or vote down the ones you don't like,
and go to the original project or source file by following the links above each example.
You may also want to check out all available functions/classes of the module
tensorflow.python.ops.gen_nn_ops
, or try the search function
.
.
Example #1
| Source File: nn_grad.py From lambda-packs with MIT License | 6 votes |
|
def _BatchNormWithGlobalNormalizationGrad(op, grad):
"""Return the gradients for the 5 inputs of BatchNormWithGlobalNormalization.
We do not backprop anything for the mean and var intentionally as they are
not being trained with backprop in the operation.
Args:
op: The BatchNormOp for which we need to generate gradients.
grad: Tensor. The gradients passed to the BatchNormOp.
Returns:
dx: Backprop for input, which is (grad * (g * rsqrt(v + epsilon)))
dm: Backprop for mean, which is
sum_over_rest(grad * g) * (-1 / rsqrt(v + epsilon))
dv: Backprop for variance, which is
sum_over_rest(grad * g * (x - m)) * (-1/2) * (v + epsilon) ^ (-3/2)
db: Backprop for beta, which is grad reduced in all except the
last dimension.
dg: Backprop for gamma, which is (grad * ((x - m) * rsqrt(v + epsilon)))
"""
dx, dm, dv, db, dg = gen_nn_ops._batch_norm_with_global_normalization_grad(
op.inputs[0], op.inputs[1], op.inputs[2], op.inputs[4], grad,
op.get_attr("variance_epsilon"), op.get_attr("scale_after_normalization"))
return dx, dm, dv, db, dg
Example #2
| Source File: nn_grad.py From auto-alt-text-lambda-api with MIT License | 6 votes |
|
def _BatchNormWithGlobalNormalizationGrad(op, grad):
"""Return the gradients for the 5 inputs of BatchNormWithGlobalNormalization.
We do not backprop anything for the mean and var intentionally as they are
not being trained with backprop in the operation.
Args:
op: The BatchNormOp for which we need to generate gradients.
grad: Tensor. The gradients passed to the BatchNormOp.
Returns:
dx: Backprop for input, which is (grad * (g * rsqrt(v + epsilon)))
dm: Backprop for mean, which is
sum_over_rest(grad * g) * (-1 / rsqrt(v + epsilon))
dv: Backprop for variance, which is
sum_over_rest(grad * g * (x - m)) * (-1/2) * (v + epsilon) ^ (-3/2)
db: Backprop for beta, which is grad reduced in all except the
last dimension.
dg: Backprop for gamma, which is (grad * ((x - m) * rsqrt(v + epsilon)))
"""
dx, dm, dv, db, dg = gen_nn_ops._batch_norm_with_global_normalization_grad(
op.inputs[0], op.inputs[1], op.inputs[2], op.inputs[4], grad,
op.get_attr("variance_epsilon"), op.get_attr("scale_after_normalization"))
return dx, dm, dv, db, dg
Example #3
| Source File: nn_grad.py From deep_image_model with Apache License 2.0 | 6 votes |
|
def _BatchNormWithGlobalNormalizationGrad(op, grad):
"""Return the gradients for the 5 inputs of BatchNormWithGlobalNormalization.
We do not backprop anything for the mean and var intentionally as they are
not being trained with backprop in the operation.
Args:
op: The BatchNormOp for which we need to generate gradients.
grad: Tensor. The gradients passed to the BatchNormOp.
Returns:
dx: Backprop for input, which is (grad * (g * rsqrt(v + epsilon)))
dm: Backprop for mean, which is
sum_over_rest(grad * g) * (-1 / rsqrt(v + epsilon))
dv: Backprop for variance, which is
sum_over_rest(grad * g * (x - m)) * (-1/2) * (v + epsilon) ^ (-3/2)
db: Backprop for beta, which is grad reduced in all except the
last dimension.
dg: Backprop for gamma, which is (grad * ((x - m) * rsqrt(v + epsilon)))
"""
dx, dm, dv, db, dg = gen_nn_ops._batch_norm_with_global_normalization_grad(
op.inputs[0], op.inputs[1], op.inputs[2], op.inputs[4], grad,
op.get_attr("variance_epsilon"), op.get_attr("scale_after_normalization"))
return dx, dm, dv, db, dg
Example #4
| Source File: nn_grad.py From Serverless-Deep-Learning-with-TensorFlow-and-AWS-Lambda with MIT License | 6 votes |
|
def _BatchNormWithGlobalNormalizationGrad(op, grad):
"""Return the gradients for the 5 inputs of BatchNormWithGlobalNormalization.
We do not backprop anything for the mean and var intentionally as they are
not being trained with backprop in the operation.
Args:
op: The BatchNormOp for which we need to generate gradients.
grad: Tensor. The gradients passed to the BatchNormOp.
Returns:
dx: Backprop for input, which is (grad * (g * rsqrt(v + epsilon)))
dm: Backprop for mean, which is
sum_over_rest(grad * g) * (-1 / rsqrt(v + epsilon))
dv: Backprop for variance, which is
sum_over_rest(grad * g * (x - m)) * (-1/2) * (v + epsilon) ^ (-3/2)
db: Backprop for beta, which is grad reduced in all except the
last dimension.
dg: Backprop for gamma, which is (grad * ((x - m) * rsqrt(v + epsilon)))
"""
dx, dm, dv, db, dg = gen_nn_ops._batch_norm_with_global_normalization_grad(
op.inputs[0], op.inputs[1], op.inputs[2], op.inputs[4], grad,
op.get_attr("variance_epsilon"), op.get_attr("scale_after_normalization"))
return dx, dm, dv, db, dg
Example #5
| Source File: nn_grad.py From keras-lambda with MIT License | 6 votes |
|
def _BatchNormWithGlobalNormalizationGrad(op, grad):
"""Return the gradients for the 5 inputs of BatchNormWithGlobalNormalization.
We do not backprop anything for the mean and var intentionally as they are
not being trained with backprop in the operation.
Args:
op: The BatchNormOp for which we need to generate gradients.
grad: Tensor. The gradients passed to the BatchNormOp.
Returns:
dx: Backprop for input, which is (grad * (g * rsqrt(v + epsilon)))
dm: Backprop for mean, which is
sum_over_rest(grad * g) * (-1 / rsqrt(v + epsilon))
dv: Backprop for variance, which is
sum_over_rest(grad * g * (x - m)) * (-1/2) * (v + epsilon) ^ (-3/2)
db: Backprop for beta, which is grad reduced in all except the
last dimension.
dg: Backprop for gamma, which is (grad * ((x - m) * rsqrt(v + epsilon)))
"""
dx, dm, dv, db, dg = gen_nn_ops._batch_norm_with_global_normalization_grad(
op.inputs[0], op.inputs[1], op.inputs[2], op.inputs[4], grad,
op.get_attr("variance_epsilon"), op.get_attr("scale_after_normalization"))
return dx, dm, dv, db, dg
Example #6
| Source File: nn_batchnorm_test.py From deep_image_model with Apache License 2.0 | 4 votes |
|
def testBatchNormGradImpl(self):
x_shape = [7, 5, 4, 6]
param_shape = [6]
np.random.seed(1) # Make it reproducible.
x_val = np.random.random_sample(x_shape).astype(np.float32)
m_val = np.random.random_sample(param_shape).astype(np.float32)
v_val = np.random.random_sample(param_shape).astype(np.float32)
beta_val = np.random.random_sample(param_shape).astype(np.float32)
gamma_val = np.random.random_sample(param_shape).astype(np.float32)
backprop_val = np.random.random_sample(x_shape).astype(np.float32)
for use_gpu in [False, True]:
with self.test_session(use_gpu=use_gpu) as sess:
x = tf.constant(x_val, name="x")
m = tf.constant(m_val, name="m")
v = tf.constant(v_val, name="v")
beta = tf.constant(beta_val, name="beta")
gamma = tf.constant(gamma_val, name="gamma")
backprop = tf.constant(backprop_val, name="backprop")
epsilon = 0.001
for scale_after_normalization in [True, False]:
# _batch_norm_with_global_normalization_grad is deprecated in v9
tf.get_default_graph().graph_def_versions.producer = 8
grad = gen_nn_ops._batch_norm_with_global_normalization_grad(
x, m, v, gamma, backprop, epsilon, scale_after_normalization)
dx, dm, dv, db, dg = grad
self.assertEqual(grad.dx, dx)
self.assertEqual(grad.dm, dm)
self.assertEqual(grad.dv, dv)
self.assertEqual(grad.db, db)
self.assertEqual(grad.dg, dg)
on = self._opsBatchNorm(
x, m, v, beta, gamma, epsilon, scale_after_normalization, True)
odx, odm, odv, odb, odg = tf.gradients(
[on], [x, m, v, beta, gamma], [backprop])
if scale_after_normalization:
all_grads = sess.run([dx, dm, dv, db, dg, odx, odm, odv, odb, odg])
to_check = ["dx", "dm", "dv", "db", "dg"]
else:
all_grads = sess.run([dx, dm, dv, db, odx, odm, odv, odb])
to_check = ["dx", "dm", "dv", "db"]
for i, _ in enumerate(to_check):
self.assertAllClose(
all_grads[i + len(to_check)], all_grads[i], atol=0.000001)
