import numpy as np
import torch
import torchvision
# a example of 2-layer neural network
# start from basic logic
N, D_IN, H, D_OUT=64,1000,100,10
x=torch.randn(N,D_IN).cuda()
y=torch.randn(N,D_OUT).cuda()
w1=torch.randn(D_IN,H,requires_grad=True).cuda()
w2=torch.randn(H,D_OUT,requires_grad=True).cuda()
learning_rate=1e-6
for it in range(500):
y_pred=x.mm(w1).clamp(min=0).mm(w2)
loss=(y_pred-y).pow(2).sum()
print(it,loss.item())
grad_w2=x.mm(w1).T.clamp(min=0).mm(2*(y_pred-y))
grad_w1=x.T.mm(2*(y_pred-y)).mm(w2.T)
w1 -=learning_rate * grad_w1
w2 -=learning_rate * grad_w2
仅用于理解梯度下降和神经网络训练的逻辑和原理。
另外最近加了显卡顺便也试一下cuda是否可用。亲测无问题,运行结果如下:(没有CUDA把.cuda()去掉就可以)
.........................................
482 7.107484270818532e-05
483 7.013617869233713e-05
484 6.890020449645817e-05