本文整理汇总了Python中tensorflow.python.ops.gen_math_ops._rsqrt_grad方法的典型用法代码示例。如果您正苦于以下问题：Python gen_math_ops._rsqrt_grad方法的具体用法？Python gen_math_ops._rsqrt_grad怎么用？Python gen_math_ops._rsqrt_grad使用的例子？那么, 这里精选的方法代码示例或许可以为您提供帮助。您也可以进一步了解该方法所在类tensorflow.python.ops.gen_math_ops的用法示例。
在下文中一共展示了gen_math_ops._rsqrt_grad方法的5个代码示例，这些例子默认根据受欢迎程度排序。您可以为喜欢或者感觉有用的代码点赞，您的评价将有助于系统推荐出更棒的Python代码示例。
示例1: _RsqrtGrad
# 需要导入模块: from tensorflow.python.ops import gen_math_ops [as 别名]
# 或者: from tensorflow.python.ops.gen_math_ops import _rsqrt_grad [as 别名]
def _RsqrtGrad(op, grad):
"""Returns -0.5 * grad * conj(y)^3."""
y = op.outputs[0] # y = x^(-1/2)
return gen_math_ops._rsqrt_grad(y, grad) 示例2: _RsqrtGradGrad
# 需要导入模块: from tensorflow.python.ops import gen_math_ops [as 别名]
# 或者: from tensorflow.python.ops.gen_math_ops import _rsqrt_grad [as 别名]
def _RsqrtGradGrad(op, grad):
"""Returns backprop gradient for f(a,b) = -0.5 * b * conj(a)^3."""
a = op.inputs[0] # a = x^{-1/2}
b = op.inputs[1] # backprop gradient for a
with ops.control_dependencies([grad.op]):
ca = math_ops.conj(a)
cg = math_ops.conj(grad)
grad_a = -1.5 * cg * b * math_ops.square(ca)
# pylint: disable=protected-access
grad_b = gen_math_ops._rsqrt_grad(ca, grad)
return grad_a, grad_b 示例3: testGradGrad
# 需要导入模块: from tensorflow.python.ops import gen_math_ops [as 别名]
# 或者: from tensorflow.python.ops.gen_math_ops import _rsqrt_grad [as 别名]
def testGradGrad(self):
np.random.seed(7)
shape = (5,)
dtype_tols = [(np.float32, 5e-4), (np.float64, 1e-6), (np.complex64, 5e-4),
(np.complex128, 1e-6)]
op_range = [(gen_math_ops._inv_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):
x = np.random.uniform(
real_range[0], real_range[1], size=shape[0]).astype(dtype)
if dtype in (np.complex64, np.complex128):
x += 1j * np.random.uniform(-2, 2, size=shape[0]).astype(dtype)
return x
for op, real_range in op_range:
with self.test_session():
for dtype, tol in dtype_tols:
x = tf.constant(rand(dtype))
y = tf.constant(rand(dtype))
z = op(x, y)
grads = tf.test.compute_gradient(
[x, y], [shape, shape],
z,
shape,
x_init_value=[rand(dtype), rand(dtype)])
if isinstance(grads, tuple):
grads = [grads]
for analytical, numerical in grads:
self.assertAllClose(analytical, numerical, rtol=tol, atol=tol) 示例4: _RsqrtGrad
# 需要导入模块: from tensorflow.python.ops import gen_math_ops [as 别名]
# 或者: from tensorflow.python.ops.gen_math_ops import _rsqrt_grad [as 别名]
def _RsqrtGrad(op, grad):
"""Returns -0.5 * grad * conj(y)^3."""
y = op.outputs[0] # y = x^(-1/2)
# pylint: disable=protected-access
return gen_math_ops._rsqrt_grad(y, grad)
# pylint: enable=protected-access 开发者ID:PacktPublishing，项目名称:Serverless-Deep-Learning-with-TensorFlow-and-AWS-Lambda，代码行数:8，代码来源:math_grad.py
示例5: _RsqrtGradGrad
# 需要导入模块: from tensorflow.python.ops import gen_math_ops [as 别名]
# 或者: from tensorflow.python.ops.gen_math_ops import _rsqrt_grad [as 别名]
def _RsqrtGradGrad(op, grad):
"""Returns backprop gradient for f(a,b) = -0.5 * b * conj(a)^3."""
a = op.inputs[0] # a = x^{-1/2}
b = op.inputs[1] # backprop gradient for a
with ops.control_dependencies([grad]):
ca = math_ops.conj(a)
cg = math_ops.conj(grad)
grad_a = -1.5 * cg * b * math_ops.square(ca)
# pylint: disable=protected-access
grad_b = gen_math_ops._rsqrt_grad(ca, grad)
return grad_a, grad_b 开发者ID:PacktPublishing，项目名称:Serverless-Deep-Learning-with-TensorFlow-and-AWS-Lambda，代码行数:13，代码来源:math_grad.py