对抗生成网络_生成对抗网络 | 实验

上期我们介绍了

生成对抗网络 | 原理及训练过程

同样地，我们依旧通过实验来巩固我们刚刚所学的知识点。本次实验是基于Jupyer Notebook、Anaconda Python3.7与Keras环境。数据集是利用Minst手写体图像数据集。

5.3.1 代码

1.  # chapter5/5_3_GAN.ipynb2.  import random  3.  import numpy as np  4.  from keras.layers import Input  5.  from keras.layers.core import Reshape,Dense,Dropout,Activation,Flatten  6.  from keras.layers.advanced_activations import LeakyReLU  7.  from keras.layers.convolutional import Convolution2D, MaxPooling2D, ZeroPadding2D, Deconv2D, UpSampling2D  8.  from keras.regularizers import *  9.  from keras.layers.normalization import *  10.  from keras.optimizers import *  11.  from keras.datasets import mnist  12.  import matplotlib.pyplot as plt  13.  from keras.models import Model  14.  from tqdm import tqdm  15.  from IPython import display  

1. 读取数据集

1.  img_rows, img_cols = 28, 28  2.    3.  # 数据集的切分与混洗(shuffle) 4.  (X_train, y_train), (X_test, y_test) = mnist.load_data()  5.    6.  X_train = X_train.reshape(X_train.shape[0], 1, img_rows, img_cols)  7.  X_test = X_test.reshape(X_test.shape[0], 1, img_rows, img_cols)  8.  X_train = X_train.astype('float32')  9.  X_test = X_test.astype('float32')  10.  X_train /= 255  11.  X_test /= 255  12.    13.  print(np.min(X_train), np.max(X_train))  14.  print('X_train shape:', X_train.shape)  15.  print(X_train.shape[0], 'train samples')  16.  print(X_test.shape[0], 'test samples')  

0.0 1.0

X_train shape: (60000, 1, 28,28)

60000 train samples

10000 test samples

2. 超参数设置

1.  shp = X_train.shape[1:]  2.  dropout_rate = 0.25  3.    4.  # 优化器  5.  opt = Adam(lr=1e-4)