1 import torch
2 import torch.nn.functional as F
3 import matplotlib.pyplot as plt
4
5 # torch.manual_seed(1) # reproducible
6
7 x = torch.unsqueeze(torch.linspace(-1, 1, 100), dim=1) # x data (tensor), shape=(100, 1)
8 y = x.pow(2) + 0.2*torch.rand(x.size()) # noisy y data (tensor), shape=(100, 1)
9
10 # torch can only train on Variable, so convert them to Variable
11 # The code below is deprecated in Pytorch 0.4. Now, autograd directly supports tensors
12 # x, y = Variable(x), Variable(y)
13
14 # plt.scatter(x.data.numpy(), y.data.numpy())
15 # plt.show()
16
17
18 class Net(torch.nn.Module):
19 def __init__(self, n_feature, n_hidden, n_output):
20 super(Net, self).__init__()
21 self.hidden = torch.nn.Linear(n_feature, n_hidden) # hidden layer
22 self.predict = torch.nn.Linear(n_hidden, n_output) # output layer
23
24 def forward(self, x):
25 x = F.relu(self.hidden(x)) # activation function for hidden layer
26 x = self.predict(x) # linear output
27 return x
28
29 net = Net(n_feature=1, n_hidden=10, n_output=1) # define the network
30 print(net) # net architecture
31
32 optimizer = torch.optim.SGD(net.parameters(), lr=0.2)
33 loss_func = torch.nn.MSELoss() # this is for regression mean squared loss
34
35 plt.ion() # something about plotting
36
37 for t in range(200):
38 prediction = net(x) # input x and predict based on x
39
40 loss = loss_func(prediction, y) # must be (1. nn output, 2. target)
41
42 optimizer.zero_grad() # clear gradients for next train
43 loss.backward() # backpropagation, compute gradients
44 optimizer.step() # apply gradients
45
46 if t % 5 == 0:
47 # plot and show learning process
48 plt.cla()
49 plt.scatter(x.data.numpy(), y.data.numpy())
50 plt.plot(x.data.numpy(), prediction.data.numpy(), 'r-', lw=5)
51 plt.text(0.5, 0, 'Loss=%.4f' % loss.data.numpy(), fontdict={'size': 20, 'color': 'red'})
52 plt.pause(0.1)
53
54 plt.ioff()
55 plt.show()
