|
import torch as t
|
|
import torch.nn as nn
|
|
|
|
x = t.randn(4,3)
|
|
y = t.randn(4,2)
|
|
|
|
linear = nn.Linear(3,2)
|
|
|
|
print('w: ', linear.weight)
|
|
print('b: ', linear.bias)
|
|
|
|
criterion = nn.MSELoss()
|
|
optimizer = t.optim.SGD(linear.parameters(), lr=0.01)
|
|
|
|
pred = linear(x)
|
|
|
|
loss = criterion(pred, y)
|
|
print('loss: ', loss.item())
|
|
|
|
loss.backward()
|
|
|
|
print('dL/dw: ', linear.weight.grad)
|
|
print('dL/db: ', linear.bias.grad)
|
|
|
|
|
|
optimizer.step()
|
|
|
|
pred = linear(x)
|
|
loss = criterion(pred, y)
|
|
|
|
print('loss after 1 step optimization', loss.item()) |