画loss曲线：

写在运行代码里面：
步骤一：
 

train_losses = []
train_losses.append(round(float(loss.data.cpu()),3))

步骤二： 

torch.save(train_losses,  '/home/wu/local/loss.pth')

保存loss

独立运行：

import torch
import matplotlib.pyplot as plt
import numpy as np

loss = torch.load('./loss.pth')
num = len(loss)
x = [i+1 for i in range(num)]
print(num)
plt.figure(1)
plt.plot(loss[:])
plt.show()