from mxnet import nd from mxnet import autograd import math x=nd.array([[10,20],[1,2]]) x.attach_grad()#在ndarray里準備存儲梯度 with autograd.record():c=nd.log(x) #反向傳播backward() c.backward() #f'(x)=1/x z=x.grad print(x) print(z)

[[10. 20.]
[ 1. 2.]]
<NDArray 2x2 @cpu(0)>

[[0.1 0.05]
[1. 0.5 ]]
<NDArray 2x2 @cpu(0)>

from mxnet import nd from mxnet import autograd import math x1=nd.array([[1,2],[3,4]]) x2=nd.array([[10,20],[30,40]]) x1.attach_grad()#在ndarray里準備存儲梯度 x2.attach_grad()#在ndarray里準備存儲梯度 def f(x): if x.sum()>20:y=3*x*xelse:y=nd.log(x)return y with autograd.record():c1=f(x1)c2=f(x2) #反向傳播backward() c1.backward() c2.backward() #f'(x)=1/x or f'(x)=3*2*x z1=x1.grad z2=x2.grad print(x1) print(z1) print(x2) print(z2)

[[1. 2.]
[3. 4.]]
<NDArray 2x2 @cpu(0)>

[[1. 0.5 ]
[0.33333334 0.25 ]]
<NDArray 2x2 @cpu(0)>

[[10. 20.]
[30. 40.]]
<NDArray 2x2 @cpu(0)>

[[ 60. 120.]
[180. 240.]]
<NDArray 2x2 @cpu(0)>

總結(jié)

以上是生活随笔為你收集整理的mxnet基础到提高(38)-自动梯度计算的全部內(nèi)容，希望文章能夠幫你解決所遇到的問題。

如果覺得生活随笔網(wǎng)站內(nèi)容還不錯，歡迎將生活随笔推薦給好友。