-
Notifications
You must be signed in to change notification settings - Fork 0
/
train.py
159 lines (127 loc) · 5.6 KB
/
train.py
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
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
# -*- coding: utf-8 -*-
from __future__ import print_function
import argparse
import matplotlib.pyplot as plt
import numpy as np
import chainer
from chainer import computational_graph
from chainer import cuda
from chainer import optimizers,optimizer
from chainer import serializers
from chainer import functions as F
import os, sys, time
from PIL import Image
import utils
from model import Discriminator, AAE
parser = argparse.ArgumentParser(description='Chainer example: MNIST')
parser.add_argument('--initmodel', '-m', default='',
help='Initialize the model from given file')
parser.add_argument('--resume', '-r', default='',
help='Resume the optimization from snapshot')
parser.add_argument('--gpu', '-g', default=-1, type=int,
help='GPU ID (negative value indicates CPU)')
parser.add_argument('--epoch', '-e', default=100, type=int,
help='number of epochs to learn')
parser.add_argument('--dimz', '-z', default=2, type=int,
help='dimention of encoded vector')
parser.add_argument('--batchsize', '-b', type=int, default=100,
help='learning minibatch size')
parser.add_argument("--data_dir", type=str, default="./data")
parser.add_argument("--model_dir", type=str, default="./models")
parser.add_argument("--visualization_dir", type=str, default="./visualization")
parser.add_argument("--load_epoch", type=int, default=0)
args = parser.parse_args()
batchsize = args.batchsize
n_epoch = args.epoch
n_z = args.dimz
if n_z % 2 != 0:
raise Exception("The dimension of the latent code z must be a multiple of 2.")
print('GPU: {}'.format(args.gpu))
print('# dim z: {}'.format(args.dimz))
print('# Minibatch-size: {}'.format(args.batchsize))
print('# epoch: {}'.format(args.epoch))
try:
os.mkdir(args.model_dir)
except:
pass
try:
os.mkdir(args.visualization_dir)
except:
pass
x_train, x_test, y_train, y_test = utils.load_mnist()
N = len(x_train)
model = AAE(784, n_z, hidden_units_enc=(1000, 1000, 500), hidden_units_dec=(500,1000,1000))
dis = Discriminator(n_z+10)
use_gpu = args.gpu >= 0
if use_gpu:
cuda.get_device(args.gpu).use()
model.to_gpu()
dis.to_gpu()
xp = np if args.gpu < 0 else cuda.cupy
optimizer_dis = optimizers.Adam(alpha=0.0002, beta1=0.5)
optimizer_aae = optimizers.Adam(alpha=0.0002, beta1=0.5)
optimizer_dis.setup(dis)
optimizer_aae.setup(model)
optimizer_dis.add_hook(optimizer.WeightDecay(0.0001))
optimizer_aae.add_hook(optimizer.WeightDecay(0.0001))
n_steps_to_optimize_dis = 1
total_time = 0
C = 10
for epoch in range(1, n_epoch+1):
print('Epoch {}'.format(epoch))
sum_loss_regularization = 0
sum_loss_reconstruction = 0
start_time = time.time()
perm = np.random.permutation(N)
for i in range(0, N, batchsize):
x = chainer.Variable(xp.asarray(x_train[perm[i:i + batchsize]]))
y = chainer.Variable(xp.asarray(y_train[perm[i:i + batchsize]].astype(np.float32)))
y_label = np.argmax(y_train[perm[i:i + batchsize]], axis=1)
# Reconstruction phase
z_fake_batch = model.encode(x)
_x = model.decode(z_fake_batch, sigmoid=False)
loss_reconstruction = F.bernoulli_nll(x, _x) / batchsize
# Adversarial phase
z_real_batch = utils.sample_z_from_n_2d_gaussian_mixture(batchsize, n_z, y_label, 10, use_gpu)
z_real_batch_with_label = F.concat((z_real_batch, y))
p_real_batch = dis(z_real_batch_with_label)
z_fake_batch_with_label = F.concat((z_fake_batch, y))
p_fake_batch = dis(z_fake_batch_with_label)
loss_dis_real = F.softmax_cross_entropy(p_real_batch, chainer.Variable(xp.zeros(batchsize, dtype=np.int32)))
loss_dis_fake = F.softmax_cross_entropy(p_fake_batch, chainer.Variable(xp.ones(batchsize, dtype=np.int32)))
loss_dis = loss_dis_fake + loss_dis_real
loss_gen = F.softmax_cross_entropy(p_fake_batch, chainer.Variable(xp.zeros(batchsize, dtype=np.int32)))
loss_aae = loss_reconstruction + C * loss_gen
optimizer_dis.zero_grads()
loss_dis.backward()
optimizer_dis.update()
optimizer_aae.zero_grads()
loss_aae.backward()
optimizer_aae.update()
# sum_loss_regularization += float(loss_dis.data) * batchsize_loop
sum_loss_regularization += float(C*loss_gen.data) * batchsize
sum_loss_reconstruction += float(loss_reconstruction.data) * batchsize
# Saving the models
print(" reconstruction_loss", (sum_loss_reconstruction / N))
print(" regularization_loss", (sum_loss_regularization / N))
p_real_batch.to_cpu()
p_real_batch = p_real_batch.data.transpose(1, 0)
p_real_batch = np.exp(p_real_batch)
sum_p_real_batch = p_real_batch[0] + p_real_batch[1]
win_real = p_real_batch[0] / sum_p_real_batch
print(" D(real_z)", win_real.mean())
p_fake_batch.to_cpu()
p_fake_batch = p_fake_batch.data.transpose(1, 0)
p_fake_batch = np.exp(p_fake_batch)
sum_p_fake_batch = p_fake_batch[0] + p_fake_batch[1]
win_fake = p_fake_batch[0] / sum_p_fake_batch
print(" D(enc_z) ", win_fake.mean())
elapsed_time = time.time() - start_time
# print " time", elapsed_time
total_time += elapsed_time
print(" total_time", total_time)
if epoch % 10 == 0:
serializers.save_hdf5("%s/aae_epoch_%d.model" % (args.model_dir, epoch), model)
serializers.save_hdf5("%s/dis_epoch_%d.model" % (args.model_dir, epoch), dis)
utils.visualize_labeled_z(xp, model, x_test[:1000], np.argmax(y_test[:1000], axis=1), args.visualization_dir, epoch)
utils.visualize_reconstruction(xp, model, x_test[:50], args.visualization_dir, epoch)