對抗生成網路原理和作用
阿新 • • 發佈:2019-02-05
我們通過一個demo(gan.py )來講解對抗生成網路的原理和作用
1、建立真實資料
2、使用GAN訓練噪聲資料
3、通過1200次的訓練使得生成的資料的分佈跟真實資料的分佈差不多
4、通過debug方式一步步的講解
二、原理:
1、G(x)是生成的資料,放到判別D網路中,希望D網路輸出 0;x是真實的輸入,希望D網路輸出 1
2、x輸入G網路通過一系列的引數生成G(x)
3、對於D網路希望他的判別標準要高些,這樣生成的資料才更能接近真實資料,這就需要D_pre網路進行預先的判斷
三、程式碼實現的主要步驟:
1、構造判別網路模型 3–14
2、構造生成網路模型 15–32
3、構造損失函式 33–35
4、訓練對抗生成網路
import argparse #1、引數解析的包
import numpy as np
from scipy.stats import norm
import tensorflow as tf
import matplotlib.pyplot as plt
from matplotlib import animation
import seaborn as sns #2、視覺化的庫
sns.set(color_codes=True)
seed = 42
np.random.seed(seed)
tf.set_random_seed(seed)
class DataDistribution (object):
def __init__(self):
self.mu = 4
self.sigma = 0.5
#44、
def sample(self, N):
samples = np.random.normal(self.mu, self.sigma, N)
samples.sort()
return samples
#6、隨機初始化分佈,作為噪音點
class GeneratorDistribution(object):
def __init__(self, range) :
self.range = range
def sample(self, N):
return np.linspace(-self.range, self.range, N) + \
np.random.random(N) * 0.01
#16、
def linear(input, output_dim, scope=None, stddev=1.0):
#17、定義一個隨機的初始化
norm = tf.random_normal_initializer(stddev=stddev)
#18、初始化常量為0
const = tf.constant_initializer(0.0)
with tf.variable_scope(scope or 'linear'):
#19、w進行高斯處理話
w = tf.get_variable('w', [input.get_shape()[1], output_dim], initializer=norm)
#20、b進行常量初始化
b = tf.get_variable('b', [output_dim], initializer=const)
return tf.matmul(input, w) + b
#29、生成網路只要兩層就可以產生最終的輸出結果
def generator(input, h_dim):
h0 = tf.nn.softplus(linear(input, h_dim, 'g0'))
h1 = linear(h0, 1, 'g1')
return h1
# 15、h0~h3 是分層的
def discriminator(input, h_dim):
#h0是第一層的輸出,h_dim * 2 隱層的資料
h0 = tf.tanh(linear(input, h_dim * 2, 'd0'))
h1 = tf.tanh(linear(h0, h_dim * 2, 'd1'))
h2 = tf.tanh(linear(h1, h_dim * 2, scope='d2'))
#21、h3我們網路最總的輸出結果
h3 = tf.sigmoid(linear(h2, 1, scope='d3'))
return h3
#24、優化器,學習率不斷衰減的策略
def optimizer(loss, var_list, initial_learning_rate):
decay = 0.95
num_decay_steps = 150
batch = tf.Variable(0)
#25、學習率不斷衰減的學習方式
learning_rate = tf.train.exponential_decay(
initial_learning_rate,
batch,
num_decay_steps,
decay,
staircase=True
)
#26、通過梯度下降定義求解器
optimizer = tf.train.GradientDescentOptimizer(learning_rate).minimize(
loss,
global_step=batch,
var_list=var_list
)
return optimizer
class GAN(object):
#9、
def __init__(self, data, gen, num_steps, batch_size, log_every):
self.data = data
self.gen = gen
self.num_steps = num_steps
self.batch_size = batch_size
self.log_every = log_every
self.mlp_hidden_size = 4
self.learning_rate = 0.03
#10、
self._create_model()
def _create_model(self):
#11、構建D網路的骨架
with tf.variable_scope('D_pre'):
#12、輸入,注意shape的引數
self.pre_input = tf.placeholder(tf.float32, shape=(self.batch_size, 1))
#13、label
self.pre_labels = tf.placeholder(tf.float32, shape=(self.batch_size, 1))
#14、初始化操作
D_pre = discriminator(self.pre_input, self.mlp_hidden_size)
#22、預測值與真實值的差異D_pre和pre_labels的差異
self.pre_loss = tf.reduce_mean(tf.square(D_pre - self.pre_labels))
#23、
self.pre_opt = optimizer(self.pre_loss, None, self.learning_rate)
# This defines the generator network - it takes samples from a noise
# distribution as input, and passes them through an MLP.
with tf.variable_scope('Gen'):
#27、噪音的輸入
self.z = tf.placeholder(tf.float32, shape=(self.batch_size, 1))
#28、G網路用於資料的生成
self.G = generator(self.z, self.mlp_hidden_size)
# The discriminator tries to tell the difference between samples from the
# true data distribution (self.x) and the generated samples (self.z).
#
# Here we create two copies of the discriminator network (that share parameters),
# as you cannot use the same network with different inputs in TensorFlow.
with tf.variable_scope('Disc') as scope:
#30、D網路使用者判別功能
self.x = tf.placeholder(tf.float32, shape=(self.batch_size, 1))
#31、self.x 是真實的資料
self.D1 = discriminator(self.x, self.mlp_hidden_size)
scope.reuse_variables()
#32、self.G是生成的資料
self.D2 = discriminator(self.G, self.mlp_hidden_size)
# Define the loss for discriminator and generator networks (see the original
# paper for details), and create optimizers for both
#33、判別網路的損失函式,希望D1趨近於1,希望D2趨近於0
self.loss_d = tf.reduce_mean(-tf.log(self.D1) - tf.log(1 - self.D2))
#34、生成網路(希望騙過判別網路)的損失函式,希望loss_g趨近於1
self.loss_g = tf.reduce_mean(-tf.log(self.D2))
self.d_pre_params = tf.get_collection(tf.GraphKeys.TRAINABLE_VARIABLES, scope='D_pre')
self.d_params = tf.get_collection(tf.GraphKeys.TRAINABLE_VARIABLES, scope='Disc')
self.g_params = tf.get_collection(tf.GraphKeys.TRAINABLE_VARIABLES, scope='Gen')
#35、通過優化器不斷地優化loss_d和loss_g
self.opt_d = optimizer(self.loss_d, self.d_params, self.learning_rate)
self.opt_g = optimizer(self.loss_g, self.g_params, self.learning_rate)
#36、開始訓練
def train(self):
with tf.Session() as session:
tf.global_variables_initializer().run()
# pretraining discriminator
num_pretrain_steps = 1000
#37、先訓練D-pro
for step in range(num_pretrain_steps):
#38、
d = (np.random.random(self.batch_size) - 0.5) * 10.0
#39、
labels = norm.pdf(d, loc=self.data.mu, scale=self.data.sigma)
#40、迭代
pretrain_loss, _ = session.run([self.pre_loss, self.pre_opt], {
self.pre_input: np.reshape(d, (self.batch_size, 1)),
self.pre_labels: np.reshape(labels, (self.batch_size, 1))
})
#41、
self.weightsD = session.run(self.d_pre_params)
# 42、copy weights from pre-training over to new D network
for i, v in enumerate(self.d_params):
session.run(v.assign(self.weightsD[i]))
for step in range(self.num_steps):
# 43、update discriminator
x = self.data.sample(self.batch_size)
z = self.gen.sample(self.batch_size)
loss_d, _ = session.run([self.loss_d, self.opt_d], {
self.x: np.reshape(x, (self.batch_size, 1)),
self.z: np.reshape(z, (self.batch_size, 1))
})
# 45、迭代優化兩個網路 update generator
z = self.gen.sample(self.batch_size)
loss_g, _ = session.run([self.loss_g, self.opt_g], {
self.z: np.reshape(z, (self.batch_size, 1))
})
if step % self.log_every == 0:
print('{}: {}\t{}'.format(step, loss_d, loss_g))
if step % 100 == 0 or step==0 or step == self.num_steps -1 :
self._plot_distributions(session)
def _samples(self, session, num_points=10000, num_bins=100):
xs = np.linspace(-self.gen.range, self.gen.range, num_points)
bins = np.linspace(-self.gen.range, self.gen.range, num_bins)
# data distribution
d = self.data.sample(num_points)
pd, _ = np.histogram(d, bins=bins, density=True)
# generated samples
zs = np.linspace(-self.gen.range, self.gen.range, num_points)
g = np.zeros((num_points, 1))
for i in range(num_points // self.batch_size):
g[self.batch_size * i:self.batch_size * (i + 1)] = session.run(self.G, {
self.z: np.reshape(
zs[self.batch_size * i:self.batch_size * (i + 1)],
(self.batch_size, 1)
)
})
pg, _ = np.histogram(g, bins=bins, density=True)
return pd, pg
def _plot_distributions(self, session):
pd, pg = self._samples(session)
p_x = np.linspace(-self.gen.range, self.gen.range, len(pd))
f, ax = plt.subplots(1)
ax.set_ylim(0, 1)
plt.plot(p_x, pd, label='real data')
plt.plot(p_x, pg, label='generated data')
plt.title('1D Generative Adversarial Network')
plt.xlabel('Data values')
plt.ylabel('Probability density')
plt.legend()
plt.show()
def main(args): #3、夠造一個model
model = GAN(
#4、引數
DataDistribution(),
#5、
GeneratorDistribution(range=8),
#7、定義引數
args.num_steps,
args.batch_size,
#8、隔多長時間
args.log_every,
)
model.train()
def parse_args():
parser = argparse.ArgumentParser()
parser.add_argument('--num-steps', type=int, default=1200,
help='the number of training steps to take')
parser.add_argument('--batch-size', type=int, default=12,
help='the batch size')
parser.add_argument('--log-every', type=int, default=10,
help='print loss after this many steps')
return parser.parse_args()
if __name__ == '__main__':
main(parse_args())