-
Notifications
You must be signed in to change notification settings - Fork 10
/
discrete_mnist.py
157 lines (117 loc) · 5.52 KB
/
discrete_mnist.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
'''GAN with RWS.
'''
from collections import OrderedDict
import cortex
from cortex import set_experiment
from cortex.training.parsers import make_argument_parser
from cortex.utils.maths import norm_exp
from cortex import _manager as manager
from theano import tensor as T
def reweighted_MLE(G=None, G_samples=None, cells=None):
if G is None:
raise TypeError('Generator distribution (G) must be provided.')
g_name, d_name = cells
if isinstance(G, str):
G_cell = manager.cells[G]
G = G_cell.get_prob(*G_cell.get_params())
else:
G_cell = manager.cells[g_name]
D_cell = manager.cells[d_name]
d = D_cell(G_samples)['P']
log_py_h1 = -D_cell.neg_log_prob(1., P=d)
log_py_h0 = -D_cell.neg_log_prob(0., P=d)
log_gx = -G_cell.neg_log_prob(G_samples, P=G[None, :, :])
log_p = log_py_h1 - log_py_h0
w = T.exp(log_p)
w_tilde = norm_exp(log_p)
cost = -(w_tilde * log_gx).sum(0).mean()
return OrderedDict(cost=cost, constants=[w_tilde])
def main(name='airrws', data='MNIST', batch_size=100, dim_in=200,
n_posterior_samples=20, test=False):
if data == 'MNIST':
source = '$data/basic/mnist_binarized_salakhutdinov.pkl.gz'
distribution_type = 'binomial'
dim = 28 * 28
greyscale = True
elif data == 'CIFAR':
greyscale = True
source='$data/basic/cifar-10-batches-py/'
distribution_type = 'gaussian_unit_variance'
dim = 32 * 32
if not greyscale: dim *= 3
d_dropout = 0.
d2_dropout = 0.2
cortex.set_path(name)
# DATA ---------------------------------------------------------------------
cortex.prepare_data(data, mode='train', source=source, name='data',
greyscale=greyscale)
cortex.prepare_data(data, mode='valid', source=source, name='data',
greyscale=greyscale)
cortex.prepare_data(data, mode='test', source=source, name='data',
greyscale=greyscale)
# CELLS --------------------------------------------------------------------
# Generative model
cortex.prepare_cell('gaussian', name='noise', dim=dim_in)
cortex.prepare_cell('DistributionMLP', name='generator', dim_hs=[500, 500, 500],
h_act='softplus', batch_normalization=True, dim=dim,
weight_normalization=False, bn_mean_only=False,
distribution_type=distribution_type)
# Discriminator
cortex.prepare_cell('DistributionMLP', name='discriminator',
distribution_type='binomial',
dim=1, dropout=d_dropout,
dim_in=dim, dim_hs=[500, 200], h_act='softplus')
# GRAPH --------------------------------------------------------------------
cortex.add_step('discriminator', 'data.input', name='real')
cortex.prepare_samples('noise', batch_size)
cortex.add_step('generator', 'noise.samples', constants=['noise.samples'])
cortex.prepare_samples('generator.P', n_posterior_samples)
cortex.add_step('discriminator', 'generator.samples', name='fake',
constants=['generator.samples'])
cortex.add_step('discriminator._cost', P='fake.P', X=0., name='fake_cost')
cortex.add_step('discriminator._cost', P='real.P', X=1., name='real_cost')
cortex.add_step('noise.grid2d', random_idx=True, name='noise_grid')
cortex.add_step('generator', 'noise_grid.output', name='gen_grid')
cortex.build()
#cortex.add_cost('l2_decay', 0.002, 'discriminator.mlp.weights')
cortex.add_cost('l2_decay', 0.002, 'generator.mlp.weights')
cortex.add_cost(lambda x, y: x + y, 'fake_cost.output', 'real_cost.output',
name='discriminator_cost')
cortex.add_cost(reweighted_MLE, G='generator.P',
G_samples='generator.samples',
cells=['generator', 'discriminator'],
name='generator_cost')
train_session = cortex.create_session()
cortex.build_session(test=test)
trainer = cortex.setup_trainer(
train_session,
optimizer='sgd',
epochs=3000,
learning_rate=0.01,
batch_size=batch_size,
excludes=['noise.mu', 'noise.log_sigma'])
model_costs = [
(['discriminator.mlp', 'discriminator.distribution'], 'discriminator_cost'),
(['generator.mlp', 'generator.distribution'], 'generator_cost')]
trainer.set_optimizer(*model_costs)
valid_session = cortex.create_session(noise=False)
cortex.build_session()
evaluator = cortex.setup_evaluator(
valid_session,
valid_stat='generator_cost',
batch_size=batch_size)
monitor = cortex.setup_monitor(valid_session, modes=['train', 'valid'])
visualizer = cortex.setup_visualizer(valid_session, batch_size=100)
visualizer.add('data.viz', X='generator.P_center', name='AIRGAN_gen')
visualizer.add('data.viz', X='gen_grid.P_center', name='AIRGAN_grid')
visualizer.add('data.viz', X='data.input', name='data_sample')
cortex.train(eval_every=10, archive_every=100)
if __name__ == '__main__':
parser = make_argument_parser()
parser.add_argument('-b', '--batch_size', type=int, default=100)
parser.add_argument('-D', '--dim_in', type=int, default=200)
parser.add_argument('-p', '--n_posterior_samples', type=int, default=20)
parser.add_argument('-d', '--data', type=str, default='MNIST')
args = parser.parse_args()
kwargs = set_experiment(args)
main(**kwargs)