如何使用autograd查找MIN / MAX点

Question

假设我们有一个简单的函数y = sin（x ** 2），如何使用autograd查找所有一阶导数值为0的X：s？

Answer 1

Below code can find the point where the first derivative is zero. However, depending on random initialization it will only find one point. If you want to find all the points, you can try iterating over a lot of random initialization on some desired grid.

import torch 
import numpy as np
# initialization
x = torch.tensor(np.random.rand(1)).requires_grad_(True)

while (x.grad is None or torch.abs(x.grad)>0.01):
    if (x.grad is not None):
        # zero grads
        x.grad.data.zero_()
    # compute fn
    y = torch.sin(x**2)
    # compute grads
    y.backward()
    # move in direction of / opposite to grads
    x.data = x.data - 0.01*x.grad.data
    # use below line to move uphill 
    # x.data = x.data + 0.01*x.grad.data

print(x)
print(y)
print(x.grad)

Also see how to apply gradients manually in pytorch