InfoGAN的简易实现

这里求最大化互信息没有共享D网络，直接使用了一个简单的mlp神经网络Q

import os, sys
sys.path.append("/home/hxj/anaconda3/lib/python3.6/site-packages")
import torch
import torch.nn.functional as nn
import torch.autograd as autograd
import torch.optim as optim
import numpy as np
import matplotlib.pyplot as plt
import matplotlib.gridspec as gridspec
import os
from torch.autograd import Variable
from tensorflow.examples.tutorials.mnist import input_data


mnist = input_data.read_data_sets('./MNIST_data', one_hot=True)
mb_size = 32
Z_dim = 16
X_dim = mnist.train.images.shape[1] #784
y_dim = mnist.train.labels.shape[1] #10
h_dim = 128
cnt = 0
lr = 1e-3


def xavier_init(size):
    in_dim = size[0]
    xavier_stddev = 1. / np.sqrt(in_dim / 2.)
    return Variable(torch.randn(*size) * xavier_stddev, requires_grad=True)


""" ==================== GENERATOR ======================== """

Wzh = xavier_init(size=[Z_dim + 10, h_dim]) #shape 26 * 128
bzh = Variable(torch.zeros(h_dim), requires_grad=True)

Whx = xavier_init(size=[h_dim, X_dim]) #shape 128 * 784
bhx = Variable(torch.zeros(X_dim), requires_grad=True)


def G(z, c):
    inputs = torch.cat([z, c], 1)
    h = nn.relu(inputs @ Wzh + bzh.repeat(inputs.size(0), 1))
    X = nn.sigmoid(h @ Whx + bhx.repeat(h.size(0), 1))
    return X


""" ==================== DISCRIMINATOR ======================== """

Wxh = xavier_init(size=[X_dim, h_dim])
bxh = Variable(torch.zeros(h_dim), requires_grad=True)

Why = xavier_init(size=[h_dim, 1])
bhy = Variable(torch.zeros(1), requires_grad=True)


def D(X):
    h = nn.relu(X @ Wxh + bxh.repeat(X.size(0), 1))
    y = nn.sigmoid(h @ Why + bhy.repeat(h.size(0), 1))
    return y


""" ====================== Q(c|X) ========================== """

Wqxh = xavier_init(size=[X_dim, h_dim])
bqxh = Variable(torch.zeros(h_dim), requires_grad=True)

Whc = xavier_init(size=[h_dim, 10])
bhc = Variable(torch.zeros(10), requires_grad=True)


def Q(X):
    h = nn.relu(X @ Wqxh + bqxh.repeat(X.size(0), 1))
    c = nn.softmax(h @ Whc + bhc.repeat(h.size(0), 1))
    return c


G_params = [Wzh, bzh, Whx, bhx]
D_params = [Wxh, bxh, Why, bhy]
Q_params = [Wqxh, bqxh, Whc, bhc]
params = G_params + D_params + Q_params


""" ===================== TRAINING ======================== """


def reset_grad():
    for p in params:
        if p.grad is not None:
            data = p.grad.data
            p.grad = Variable(data.new().resize_as_(data).zero_())


G_solver = optim.Adam(G_params, lr=1e-3)
D_solver = optim.Adam(D_params, lr=1e-3)
Q_solver = optim.Adam(G_params + Q_params, lr=1e-3)


def sample_c(size):
    c = np.random.multinomial(1, 10*[0.1], size=size)
    c = Variable(torch.from_numpy(c.astype('float32')))
    return c


for it in range(100000):
    # Sample data
    X, _ = mnist.train.next_batch(mb_size) # 32
    X = Variable(torch.from_numpy(X)) #将数组转换为列向量 32*784
    z = Variable(torch.randn(mb_size, Z_dim))# 32 16 随机二维数组
    c = sample_c(mb_size) # 32 10的标签 随机标签
    print(z.shape)
    print(c.shape)
    sys.exit()

    # Dicriminator forward-loss-backward-update
    G_sample = G(z, c)
    D_real = D(X)
    D_fake = D(G_sample)

    D_loss = -torch.mean(torch.log(D_real + 1e-8) + torch.log(1 - D_fake + 1e-8))

    D_loss.backward()
    D_solver.step()

    # Housekeeping - reset gradient
    reset_grad()

    # Generator forward-loss-backward-update
    G_sample = G(z, c)
    D_fake = D(G_sample)

    G_loss = -torch.mean(torch.log(D_fake + 1e-8))

    G_loss.backward()
    G_solver.step()

    # Housekeeping - reset gradient
    reset_grad()

    # Q forward-loss-backward-update
    G_sample = G(z, c) #在c标签下生成的假样本，除了用来训练G和D之外，还要经过神经网络Q
    Q_c_given_x = Q(G_sample) # 让标签和经过Q生成的值之间的互信息最大

    crossent_loss = torch.mean(-torch.sum(c * torch.log(Q_c_given_x + 1e-8), dim=1))
    mi_loss = crossent_loss

    mi_loss.backward()
    Q_solver.step()

    # Housekeeping - reset gradient
    reset_grad()

    # Print and plot every now and then
    if it % 1000 == 0:
        idx = np.random.randint(0, 10)
        c = np.zeros([mb_size, 10])
        c[range(mb_size), idx] = 1
        c = Variable(torch.from_numpy(c.astype('float32')))
        samples = G(z, c).data.numpy()[:16]

        print('Iter-{}; D_loss: {}; G_loss: {}; Idx: {}'
              .format(it, D_loss.data.numpy(), G_loss.data.numpy(), idx))

        fig = plt.figure(figsize=(4, 4))
        gs = gridspec.GridSpec(4, 4)
        gs.update(wspace=0.05, hspace=0.05)

        for i, sample in enumerate(samples):
            ax = plt.subplot(gs[i])
            plt.axis('off')
            ax.set_xticklabels([])
            ax.set_yticklabels([])
            ax.set_aspect('equal')
            plt.imshow(sample.reshape(28, 28), cmap='Greys_r')

        if not os.path.exists('out/'):
            os.makedirs('out/')

        plt.savefig('out/{}.png'
                    .format(str(cnt).zfill(3)), bbox_inches='tight')
        cnt += 1
        plt.close(fig)