本文整理汇总了Python中chainer.Variable.zerograd方法的典型用法代码示例。如果您正苦于以下问题：Python Variable.zerograd方法的具体用法？Python Variable.zerograd怎么用？Python Variable.zerograd使用的例子？那么恭喜您, 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chainer.Variable的用法示例。
在下文中一共展示了Variable.zerograd方法的3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: _apply_backward
# 需要导入模块: from chainer import Variable [as 别名]
# 或者: from chainer.Variable import zerograd [as 别名]
def _apply_backward(self, x, grid, grads, use_cudnn):
x = Variable(x)
grid = Variable(grid)
y = functions.spatial_transformer_sampler(
x, grid, use_cudnn=use_cudnn)
x.zerograd()
grid.zerograd()
y.grad = grads
y.backward()
return x, grid, y示例2: update_step
# 需要导入模块: from chainer import Variable [as 别名]
# 或者: from chainer.Variable import zerograd [as 别名]
def update_step(net, images, step_size=1.5, end='inception_4c/output', jitter=32, clip=True):
offset_x, offset_y = np.random.randint(-jitter, jitter + 1, 2)
data = np.roll(np.roll(images, offset_x, -1), offset_y, -2)
x = Variable(xp.asarray(data))
x.zerograd()
dest, = net(x, outputs=[end])
objective(dest).backward()
g = cuda.to_cpu(x.grad)
data[:] += step_size / np.abs(g).mean() * g
data = np.roll(np.roll(data, -offset_x, -1), -offset_y, -2)
if clip:
bias = net.mean.reshape((1, 3, 1, 1))
data[:] = np.clip(data, -bias, 255 - bias)
return data示例3: Variable
# 需要导入模块: from chainer import Variable [as 别名]
# 或者: from chainer.Variable import zerograd [as 别名]
for batch_indexes in np.array_split(perm, num_batches):
x_batch = x_train[batch_indexes]
t_batch = t_train[batch_indexes]
x = Variable(x_batch)
t = Variable(t_batch)
# 順伝播
a_z = F.linear(x, w_1, b_1)
z = F.tanh(a_z)
a_y = F.linear(z, w_2, b_2)
loss = F.softmax_cross_entropy(a_y, t)
# 逆伝播
w_1.zerograd()
w_2.zerograd()
b_1.zerograd()
b_2.zerograd()
loss.backward(retain_grad=True)
grad_w_1 = w_1.grad
grad_w_2 = w_2.grad
grad_b_1 = b_1.grad
grad_b_2 = b_2.grad
w_1.data = w_1.data - learning_rate * grad_w_1
w_2.data = w_2.data - learning_rate * grad_w_2
b_1.data = b_1.data - learning_rate * grad_b_1
b_2.data = b_2.data - learning_rate * grad_b_2