线性回归的实现：Demo Pytorch

import torch
from torch.autograd import Variable
from visdom import Visdom

epochs = 5000
global_step = 0

# train data
x_data = Variable(torch.Tensor([[1.0], [2.0], [3.0]]))
y_data = Variable(torch.Tensor([[2.0], [4.0], [6.0]]))
x_data


class Model(torch.nn.Module):
    def __init__(self):
        super(Model, self).__init__()
        self.linear1 = torch.nn.Linear(1, 2) #[3,1] ->[3,2]
        self.linear2 = torch.nn.Linear(2, 1) #[3,2] ->[3,1]
        
    def forward(self, x):
        y_pred = self.linear1(x)#[3,1] ->[3,2]
        y_pred = self.linear2(y_pred)#[3,2] ->[3,1]
        return y_pred

model = Model()

viz = Visdom()

viz.line([0.], [0.], win='linear_loss', opts=dict(title='linear loss'))

criterion = torch.nn.MSELoss(size_average=False) # Defined loss function
optimizer = torch.optim.SGD(model.parameters(), lr=0.004) # Defined optimizer

# Training: forward, loss, backward, step
# Training loop
for epoch in range(epochs):
    # Forward pass
    y_pred = model(x_data)

    # Compute loss
    loss = criterion(y_pred, y_data)
    # Zero gradients
    optimizer.zero_grad()
    # perform backward pass
    loss.backward()
    # update weights
    optimizer.step()
    
    global_step +=1
    
    viz.line([loss.item()], [global_step], win='linear_loss', update='append')
    
    if loss.item() >=  1e-5:
        if epoch % 10==0:
            print("Epoch {}:\t[{}/{} ({:.0f}%)] \t Loss {:.6f}".format(epoch, global_step, epochs, 100. * global_step / epochs,loss.item()))
    else:
        print("Come on man！")
        break
    
    

Epoch 0:	[1/5000 (0%)] 	 Loss 140.462097
Epoch 10:	[11/5000 (0%)] 	 Loss 0.889901
Epoch 20:	[21/5000 (0%)] 	 Loss 0.594535
Epoch 30:	[31/5000 (1%)] 	 Loss 0.518817
Epoch 40:	[41/5000 (1%)] 	 Loss 0.452415
Epoch 50:	[51/5000 (1%)] 	 Loss 0.394169
Epoch 60:	[61/5000 (1%)] 	 Loss 0.343085
Epoch 70:	[71/5000 (1%)] 	 Loss 0.298305
Epoch 80:	[81/5000 (2%)] 	 Loss 0.259079
Epoch 90:	[91/5000 (2%)] 	 Loss 0.224751
Epoch 100:	[101/5000 (2%)] 	 Loss 0.194743
Epoch 110:	[111/5000 (2%)] 	 Loss 0.168544
Epoch 120:	[121/5000 (2%)] 	 Loss 0.145697
Epoch 130:	[131/5000 (3%)] 	 Loss 0.125802
Epoch 140:	[141/5000 (3%)] 	 Loss 0.108501
Epoch 150:	[151/5000 (3%)] 	 Loss 0.093476
Epoch 160:	[161/5000 (3%)] 	 Loss 0.080445
Epoch 170:	[171/5000 (3%)] 	 Loss 0.069159
Epoch 180:	[181/5000 (4%)] 	 Loss 0.059398
Epoch 190:	[191/5000 (4%)] 	 Loss 0.050966
Epoch 200:	[201/5000 (4%)] 	 Loss 0.043692
Epoch 210:	[211/5000 (4%)] 	 Loss 0.037423
Epoch 220:	[221/5000 (4%)] 	 Loss 0.032028
Epoch 230:	[231/5000 (5%)] 	 Loss 0.027389
Epoch 240:	[241/5000 (5%)] 	 Loss 0.023405
Epoch 250:	[251/5000 (5%)] 	 Loss 0.019987
Epoch 260:	[261/5000 (5%)] 	 Loss 0.017057
Epoch 270:	[271/5000 (5%)] 	 Loss 0.014548
Epoch 280:	[281/5000 (6%)] 	 Loss 0.012401
Epoch 290:	[291/5000 (6%)] 	 Loss 0.010565
Epoch 300:	[301/5000 (6%)] 	 Loss 0.008996
Epoch 310:	[311/5000 (6%)] 	 Loss 0.007657
Epoch 320:	[321/5000 (6%)] 	 Loss 0.006514
Epoch 330:	[331/5000 (7%)] 	 Loss 0.005540
Epoch 340:	[341/5000 (7%)] 	 Loss 0.004710
Epoch 350:	[351/5000 (7%)] 	 Loss 0.004002
Epoch 360:	[361/5000 (7%)] 	 Loss 0.003400
Epoch 370:	[371/5000 (7%)] 	 Loss 0.002888
Epoch 380:	[381/5000 (8%)] 	 Loss 0.002452
Epoch 390:	[391/5000 (8%)] 	 Loss 0.002081
Epoch 400:	[401/5000 (8%)] 	 Loss 0.001766
Epoch 410:	[411/5000 (8%)] 	 Loss 0.001498
Epoch 420:	[421/5000 (8%)] 	 Loss 0.001271
Epoch 430:	[431/5000 (9%)] 	 Loss 0.001078
Epoch 440:	[441/5000 (9%)] 	 Loss 0.000914
Epoch 450:	[451/5000 (9%)] 	 Loss 0.000775
Epoch 460:	[461/5000 (9%)] 	 Loss 0.000657
Epoch 470:	[471/5000 (9%)] 	 Loss 0.000557
Epoch 480:	[481/5000 (10%)] 	 Loss 0.000472
Epoch 490:	[491/5000 (10%)] 	 Loss 0.000400
Epoch 500:	[501/5000 (10%)] 	 Loss 0.000339
Epoch 510:	[511/5000 (10%)] 	 Loss 0.000287
Epoch 520:	[521/5000 (10%)] 	 Loss 0.000243
Epoch 530:	[531/5000 (11%)] 	 Loss 0.000206
Epoch 540:	[541/5000 (11%)] 	 Loss 0.000174
Epoch 550:	[551/5000 (11%)] 	 Loss 0.000148
Epoch 560:	[561/5000 (11%)] 	 Loss 0.000125
Epoch 570:	[571/5000 (11%)] 	 Loss 0.000106
Epoch 580:	[581/5000 (12%)] 	 Loss 0.000090
Epoch 590:	[591/5000 (12%)] 	 Loss 0.000076
Epoch 600:	[601/5000 (12%)] 	 Loss 0.000064
Epoch 610:	[611/5000 (12%)] 	 Loss 0.000054
Epoch 620:	[621/5000 (12%)] 	 Loss 0.000046
Epoch 630:	[631/5000 (13%)] 	 Loss 0.000039
Epoch 640:	[641/5000 (13%)] 	 Loss 0.000033
Epoch 650:	[651/5000 (13%)] 	 Loss 0.000028
Epoch 660:	[661/5000 (13%)] 	 Loss 0.000024
Epoch 670:	[671/5000 (13%)] 	 Loss 0.000020
Epoch 680:	[681/5000 (14%)] 	 Loss 0.000017
Epoch 690:	[691/5000 (14%)] 	 Loss 0.000014
Epoch 700:	[701/5000 (14%)] 	 Loss 0.000012
Epoch 710:	[711/5000 (14%)] 	 Loss 0.000010
Come on man！