QuickExperiments/tf_lsgan.py at master · MoMe36/QuickExperiments · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
import tensorflow as tf
import numpy as np
import pickle

import matplotlib.pyplot as plt
plt.style.use('ggplot')

dense = tf.layers.dense

class Loader:

	def __init__(self, path, max_points):

		self.path = path
		self.max_points = max_points

	def sample(self, batch):

		data = np.zeros((batch,784))
		index = np.random.randint(0,self.max_points,(batch))
		for i,ind in enumerate(index):
			im = pickle.load(open(self.path+'{}'.format(ind), 'rb'))
			data[i,:] = im.reshape(-1)

		data /= 255.
		data = (data - 0.5)/0.5
		return data

noise_size = 50
loader = Loader('/home/mehdi/Codes/MNIST/', 60000)
initializer = tf.truncated_normal_initializer(stddev=0.02)

def Generator(input_var, reuse = False):

	with tf.variable_scope('Generator', reuse = reuse):

		# input_var = tf.placeholder(tf.float32, shape = [None, noise_size], name = 'latent_variable')
		y_gen = tf.placeholder(tf.float32, shape = [None, 1], name = 'truth_measure')

		l1 = dense(input_var, 128, activation = tf.nn.leaky_relu, name = 'first_generator_layer', kernel_initializer = initializer)
		l2 = dense(l1, 400, activation = tf.nn.leaky_relu, name = 'second_generator_layer', kernel_initializer = initializer)
		generator = dense(l2, 784, activation = None, name = 'generator')

		return generator

def Discriminator(input_var, reuse= False):
	with tf.variable_scope('Discriminator', reuse = reuse):

		# x = tf.placeholder(tf.float32, shape = [None, 784], name = 'obs_variable')
		y_dis = tf.placeholder(tf.float32, shape = [None, 1], name = 'labels')

		l1 = dense(input_var, 400, activation = tf.nn.leaky_relu, name = 'first_discriminator_layer', kernel_initializer = initializer)
		l2 = dense(l1, 128, activation = tf.nn.leaky_relu, name = 'second_discriminator_layer', kernel_initializer = initializer)
		discriminator = dense(l2, 1, activation = None, name = 'generator')

		return discriminator


z = tf.placeholder(tf.float32, shape = [None, noise_size], name = 'latent_variable')
x = tf.placeholder(tf.float32, shape = [None, 784], name = 'real_data')

g = Generator(z)
d_real = Discriminator(x)
d_fake = Discriminator(g, reuse = True)

d_loss = tf.reduce_mean(tf.squared_difference(d_real, tf.ones_like(d_real)) + tf.pow(d_fake,2))
g_loss = tf.reduce_mean(tf.squared_difference(d_fake, tf.ones_like(d_fake)))

variables = tf.trainable_variables()
g_vars = [v for v in variables if 'Generator' in v.name]
d_vars = [v for v in variables if 'Discriminator' in v.name]


train_g = tf.train.AdamOptimizer(2e-4).minimize(g_loss, var_list = g_vars)
train_d = tf.train.AdamOptimizer(1e-4).minimize(d_loss, var_list = d_vars)


epochs = 10000
batch_size = 128
with tf.Session() as sess:

	sess.run(tf.global_variables_initializer())
	mean_loss_d = 0.
	mean_loss_g = 0.
	for epoch in range(1,epochs+1):

		# train D
		# for i in range(np.random.randint(1,3)):
		d_x = loader.sample(batch_size)
		noise = np.random.normal(0,1.,(batch_size, noise_size))


		_ , d_loss_var = sess.run([train_d, d_loss], feed_dict = {x:d_x, z:noise})

		mean_loss_d += d_loss_var

		# train G
		for i in range(np.random.randint(2,4)):

			noise = np.random.normal(0,1.,(batch_size, noise_size))
			_ , g_loss_var = sess.run([train_g, g_loss], feed_dict = {z:noise})

		mean_loss_g += g_loss_var

		if epoch% 100 == 0:
			print('\t\t\t === Epoch: {} === \n\nLoss D: {:.6f}\nLoss G: {:.6f}'.format(epoch, mean_loss_d/100., mean_loss_g/100.))
			mean_loss_d = 0.
			mean_loss_g = 0.

	over = False
	f, ax = plt.subplots()
	while not over:

		noise = np.random.normal(0,1.,(1, noise_size))
		prod = sess.run(g, feed_dict = {z:noise})
		ax.clear()

		ax.matshow(prod.reshape(28,28))
		plt.pause(0.1)
		over = input('Press o to quit')

		over = True if over == 'o' else False