import warnings warnings.filterwarnings("ignore") import numpy as np import pandas as pd from keras.layers import Dense, Activation, Flatten, Reshape from keras.layers import Conv2D, UpSampling2D, Conv2DTranspose from keras.layers import LeakyReLU, Dropout from keras.layers import BatchNormalization from keras.models import Sequential from keras.optimizers import Adam, RMSprop import matplotlib.pyplot as plt import theano class DCGAN(object): def __init__(self, img_rows=28, img_cols=28, channel=1): self.discriminator = None self.generator = None self.adversarial_model = None self.discriminator_model = None self.img_rows = img_rows self.img_cols = img_cols self.channel = channel def discriminator_nn(self, depth=64, dropout=0.4): if self.discriminator: return self.discriminator self.discriminator = Sequential() in_shape = (self.channel,self.img_rows, self.img_cols) self.discriminator.add(Conv2D(depth*1, (5, 5), strides=(2, 2), padding="same", input_shape=in_shape, data_format='channels_first')) self.discriminator.add(LeakyReLU(alpha=0.2)) self.discriminator.add(Dropout(dropout)) self.discriminator.add(Conv2D(depth*2, (5, 5), strides=(2,2), padding='same')) self.discriminator.add(LeakyReLU(alpha=0.2)) self.discriminator.add(Dropout(dropout)) self.discriminator.add(Conv2D(depth*4, (5, 5), strides=(2,2), padding='same')) self.discriminator.add(LeakyReLU(alpha=0.2)) self.discriminator.add(Dropout(dropout)) self.discriminator.add(Conv2D(depth*8, (5, 5), strides=(1,1), padding='same')) self.discriminator.add(LeakyReLU(alpha=0.2)) self.discriminator.add(Dropout(dropout)) self.discriminator.add(Flatten()) self.discriminator.add(Dense(1)) self.discriminator.add(Activation('sigmoid')) self.discriminator.summary() return self.discriminator def generator_nn(self, dropout = 0.4, dim = 7): if self.generator: return self.generator self.generator = Sequential() depth = 64+64+64+64 # In: 100 # Out: dim x dim x depth self.generator.add(Dense(dim*dim*depth, input_dim=100)) self.generator.add(BatchNormalization(momentum=0.9)) self.generator.add(Activation('relu')) self.generator.add(Reshape((depth, dim, dim))) self.generator.add(Dropout(dropout)) # In: dim x dim x depth # Out: 2*dim x 2*dim x depth/2 self.generator.add(UpSampling2D(data_format='channels_first')) self.generator.add(Conv2DTranspose(int(depth/2), (5, 5), padding='same',output_shape=(None, int(depth/2), 2*dim, 2*dim), data_format='channels_first')) self.generator.add(BatchNormalization(momentum=0.9)) self.generator.add(Activation('relu')) self.generator.add(UpSampling2D(data_format='channels_first')) self.generator.add(Conv2DTranspose(int(depth/4), (5, 5), padding='same',output_shape=(None ,int(depth/4), 4*dim, 4*dim), data_format='channels_first')) self.generator.add(BatchNormalization(momentum=0.9)) self.generator.add(Activation('relu')) self.generator.add(Conv2DTranspose(int(depth/8), (5, 5), padding='same',output_shape=(None ,int(depth/8), 4*dim, 4*dim), data_format='channels_first')) self.generator.add(BatchNormalization(momentum=0.9)) self.generator.add(Activation('relu')) # Out: 28 x 28 x 1 grayscale image [0.0,1.0] per pix self.generator.add(Conv2DTranspose(1, (5, 5), padding='same',output_shape=(None, 1, 4*dim, 4*dim), data_format='channels_first')) self.generator.add(Activation('sigmoid')) self.generator.summary() return self.generator def get_discriminator_model(self): if self.discriminator_model: return self.discriminator_model optimizer = RMSprop(lr=0.0008, clipvalue=1.0, decay=6e-8) self.discriminator_model = Sequential() self.discriminator_model.add(self.discriminator_nn()) self.discriminator_model.compile(loss='binary_crossentropy', optimizer=optimizer,\ metrics=['accuracy']) return self.discriminator_model def get_adversarial_model(self): if self.adversarial_model: return self.adversarial_model optimizer = RMSprop(lr=0.0004, clipvalue=1.0, decay=3e-8) self.adversarial_model = Sequential() self.adversarial_model.add(self.generator_nn()) self.adversarial_model.add(self.discriminator_nn()) self.adversarial_model.compile(loss='binary_crossentropy', optimizer=optimizer,\ metrics=['accuracy']) return self.adversarial_model class MNIST_DCGAN(object): def __init__(self): self.img_rows = 28 self.img_cols = 28 self.channel = 1 self.x_train = pd.read_csv("../input/train.csv").values self.x_train = self.x_train[:, 1:].reshape(self.x_train.shape[0], 1, self.img_rows, self.img_cols).astype(np.float32) print('train shape:', self.x_train.shape) self.DCGAN = DCGAN() self.discriminator = self.DCGAN.get_discriminator_model() self.adversarial = self.DCGAN.get_adversarial_model() self.generator = self.DCGAN.generator_nn() def train(self, train_steps=2000, batch_size=256, save_interval=0): noise_input = None if save_interval>0: noise_input = np.random.uniform(-1.0, 1.0, size=[16, 100]) for i in range(train_steps): images_train = self.x_train[np.random.randint(0, self.x_train.shape[0], size=batch_size), :, :, :] noise = np.random.uniform(-1.0, 1.0, size=[batch_size, 100]) images_fake = self.generator.predict(noise) x = np.concatenate((images_train, images_fake)) y = np.ones([2*batch_size, 1]) y[batch_size:, :] = 0 d_loss = self.discriminator.train_on_batch(x, y) y = np.ones([batch_size, 1]) noise = np.random.uniform(-1.0, 1.0, size=[batch_size, 100]) a_loss = self.adversarial.train_on_batch(noise, y) log_mesg = "step %d: [D loss: %f, acc: %f] [A loss: %f, acc: %f]" % (i, d_loss[0], d_loss[1], a_loss[0], a_loss[1]) print(log_mesg) if save_interval>0: if (i+1)%save_interval==0: self.plot_images(save2file=True, samples=noise_input.shape[0],\ noise=noise_input, step=(i+1)) def plot_images(self, save2file=False, fake=True, samples=16, noise=None, step=0): if fake: filename = 'generated_mnist.png' if noise is None: noise = np.random.uniform(-1.0, 1.0, size=[samples, 100]) else: filename = "generated_mnist_%d.png" % step images = self.generator.predict(noise) else: filename = 'real_mnist.png' i = np.random.randint(0, self.x_train.shape[0], samples) images = self.x_train[i, :, :, :] plt.figure(figsize=(10,10)) for i in range(images.shape[0]): plt.subplot(4, 4, i+1) image = images[i, :, :, :] image = np.reshape(image, [self.img_rows, self.img_cols]) plt.imshow(image, cmap='gray') plt.axis('off') plt.tight_layout() if save2file: plt.savefig(filename) plt.close('all') else: plt.show() if __name__ == '__main__': mnist_dcgan = MNIST_DCGAN() mnist_dcgan.train(train_steps=20, batch_size=256, save_interval=1) mnist_dcgan.plot_images(fake=True, save2file=True) # fake final plots mnist_dcgan.plot_images(fake=False, save2file=True) # compare with real plots