pytorch:实现简单的GAN示例(MNIST数据集)

我就废话不多说了，直接上代码吧！

# -*- coding: utf-8 -*-
"""
Created on Sat Oct 13 10:22:45 2018
@author: www
"""

import torch
from torch import nn
from torch.autograd import Variable

import torchvision.transforms as tfs
from torch.utils.data import DataLoader, sampler
from torchvision.datasets import MNIST

import numpy as np

import matplotlib.pyplot as plt
import matplotlib.gridspec as gridspec

plt.rcParams['figure.figsize'] = (10.0, 8.0) # 设置画图的尺寸
plt.rcParams['image.interpolation'] = 'nearest'
plt.rcParams['image.cmap'] = 'gray'

def show_images(images): # 定义画图工具
 images = np.reshape(images, [images.shape[0], -1])
 sqrtn = int(np.ceil(np.sqrt(images.shape[0])))
 sqrtimg = int(np.ceil(np.sqrt(images.shape[1])))

fig = plt.figure(figsize=(sqrtn, sqrtn))
 gs = gridspec.GridSpec(sqrtn, sqrtn)
 gs.update(wspace=0.05, hspace=0.05)

for i, img in enumerate(images):
   ax = plt.subplot(gs[i])
   plt.axis('off')
   ax.set_xticklabels([])
   ax.set_yticklabels([])
   ax.set_aspect('equal')
   plt.imshow(img.reshape([sqrtimg,sqrtimg]))
 return

def preprocess_img(x):
 x = tfs.ToTensor()(x)
 return (x - 0.5) / 0.5

def deprocess_img(x):
 return (x + 1.0) / 2.0

class ChunkSampler(sampler.Sampler): # 定义一个取样的函数
 """Samples elements sequentially from some offset.
 Arguments:
   num_samples: # of desired datapoints
   start: offset where we should start selecting from
 """
 def __init__(self, num_samples, start=0):
   self.num_samples = num_samples
   self.start = start

def __iter__(self):
   return iter(range(self.start, self.start + self.num_samples))

def __len__(self):
   return self.num_samples

NUM_TRAIN = 50000
NUM_VAL = 5000

NOISE_DIM = 96
batch_size = 128

train_set = MNIST('E:/data', train=True, transform=preprocess_img)

train_data = DataLoader(train_set, batch_size=batch_size, sampler=ChunkSampler(NUM_TRAIN, 0))

val_set = MNIST('E:/data', train=True, transform=preprocess_img)

val_data = DataLoader(val_set, batch_size=batch_size, sampler=ChunkSampler(NUM_VAL, NUM_TRAIN))

imgs = deprocess_img(train_data.__iter__().next()[0].view(batch_size, 784)).numpy().squeeze() # 可视化图片效果
show_images(imgs)

#判别网络
def discriminator():
 net = nn.Sequential(    
     nn.Linear(784, 256),
     nn.LeakyReLU(0.2),
     nn.Linear(256, 256),
     nn.LeakyReLU(0.2),
     nn.Linear(256, 1)
   )
 return net

#生成网络
def generator(noise_dim=NOISE_DIM):  
 net = nn.Sequential(
   nn.Linear(noise_dim, 1024),
   nn.ReLU(True),
   nn.Linear(1024, 1024),
   nn.ReLU(True),
   nn.Linear(1024, 784),
   nn.Tanh()
 )
 return net

#判别器的 loss 就是将真实数据的得分判断为 1，假的数据的得分判断为 0，而生成器的 loss 就是将假的数据判断为 1

bce_loss = nn.BCEWithLogitsLoss()#交叉熵损失函数

def discriminator_loss(logits_real, logits_fake): # 判别器的 loss
 size = logits_real.shape[0]
 true_labels = Variable(torch.ones(size, 1)).float()
 false_labels = Variable(torch.zeros(size, 1)).float()
 loss = bce_loss(logits_real, true_labels) + bce_loss(logits_fake, false_labels)
 return loss

def generator_loss(logits_fake): # 生成器的 loss
 size = logits_fake.shape[0]
 true_labels = Variable(torch.ones(size, 1)).float()
 loss = bce_loss(logits_fake, true_labels)
 return loss

# 使用 adam 来进行训练，学习率是 3e-4, beta1 是 0.5, beta2 是 0.999
def get_optimizer(net):
 optimizer = torch.optim.Adam(net.parameters(), lr=3e-4, betas=(0.5, 0.999))
 return optimizer

def train_a_gan(D_net, G_net, D_optimizer, G_optimizer, discriminator_loss, generator_loss, show_every=250,
       noise_size=96, num_epochs=10):
 iter_count = 0
 for epoch in range(num_epochs):
   for x, _ in train_data:
     bs = x.shape[0]
     # 判别网络
     real_data = Variable(x).view(bs, -1) # 真实数据
     logits_real = D_net(real_data) # 判别网络得分

sample_noise = (torch.rand(bs, noise_size) - 0.5) / 0.5 # -1 ~ 1 的均匀分布
     g_fake_seed = Variable(sample_noise)
     fake_images = G_net(g_fake_seed) # 生成的假的数据
     logits_fake = D_net(fake_images) # 判别网络得分

d_total_error = discriminator_loss(logits_real, logits_fake) # 判别器的 loss
     D_optimizer.zero_grad()
     d_total_error.backward()
     D_optimizer.step() # 优化判别网络

# 生成网络
     g_fake_seed = Variable(sample_noise)
     fake_images = G_net(g_fake_seed) # 生成的假的数据

gen_logits_fake = D_net(fake_images)
     g_error = generator_loss(gen_logits_fake) # 生成网络的 loss
     G_optimizer.zero_grad()
     g_error.backward()
     G_optimizer.step() # 优化生成网络

if (iter_count ％ show_every == 0):
       print('Iter: {}, D: {:.4}, G:{:.4}'.format(iter_count, d_total_error.item(), g_error.item()))
       imgs_numpy = deprocess_img(fake_images.data.cpu().numpy())
       show_images(imgs_numpy[0:16])
       plt.show()
       print()
     iter_count += 1

D = discriminator()
G = generator()

D_optim = get_optimizer(D)
G_optim = get_optimizer(G)

train_a_gan(D, G, D_optim, G_optim, discriminator_loss, generator_loss)      

来源：https://blog.csdn.net/xckkcxxck/article/details/83037025