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
|
import utils
import utils_blocks as ub
import theano
from theano import tensor as tsr
from blocks import algorithms, main_loop
import blocks.extensions as be
import blocks.extensions.monitoring as bm
import theano.tensor.shared_randomstreams
import numpy as np
class ClippedParams(algorithms.StepRule):
"""A rule to maintain parameters within a specified range"""
def __init__(self, min_value, max_value):
self.min_value = min_value
self.max_value = max_value
def compute_step(self, parameter, previous_step):
min_clipped = tsr.switch(parameter - previous_step < self.min_value,
0, previous_step)
return tsr.switch(parameter - previous_step > self.max_value,
0, min_clipped), []
def create_vi_model(n_nodes, n_samp=100):
"""return variational inference theano computation graph"""
def aux(a, b):
rand = a + b * np.random.normal(size=(n_nodes, n_nodes))
return np.clip(rand, 1e-3, 1 - 1e-3).astype(theano.config.floatX)
x = tsr.matrix(name='x', dtype='int8')
s = tsr.matrix(name='s', dtype='int8')
mu = theano.shared(value=aux(.5, .1), name='mu1')
sig = theano.shared(value=aux(.5, .1), name='sig1')
mu0 = theano.shared(value=aux(.5, .1), name='mu0')
sig0 = theano.shared(value=aux(.5, .1), name='sig0')
srng = tsr.shared_randomstreams.RandomStreams(seed=123)
theta = (srng.normal((n_samp, n_nodes, n_nodes)) * sig[None, :, :]
+ mu[None, :, :])
y = tsr.maximum(tsr.dot(x, theta), 1e-3)
infect = tsr.log(1. - tsr.exp(-y[0:-1])).dimshuffle(1, 0, 2)
lkl_pos = tsr.sum(infect * (x[1:] & s[1:])) / n_samp
lkl_neg = tsr.sum(-y[0:-1].dimshuffle(1, 0, 2) * (~x[1:] & s[1:])) / n_samp
lkl = lkl_pos + lkl_neg
kl = tsr.sum(tsr.log(sig0 / sig) + (sig**2 + (mu0 - mu)**2)/(2*sig0)**2)
cost = - lkl + kl
cost.name = 'cost'
return x, s, mu, sig, cost
if __name__ == "__main__":
#n_cascades = 10000
batch_size = 10
n_samples = 50
graph = utils.create_random_graph(n_nodes=4)
print('GRAPH:\n', graph, '\n-------------\n')
x, s, mu, sig, cost = create_vi_model(len(graph), n_samples)
rmse = ub.rmse_error(graph, mu)
step_rules = algorithms.CompositeRule([algorithms.AdaDelta(),
ClippedParams(1e-3, 1 - 1e-3)])
alg = algorithms.GradientDescent(cost=cost, parameters=[mu, sig],
step_rule=step_rules)
data_stream = ub.fixed_data_stream(n_cascades, graph, batch_size,
shuffle=False)
# data_stream = ub.dynamic_data_stream(graph, batch_size)
loop = main_loop.MainLoop(
alg, data_stream,
extensions=[
be.FinishAfter(after_n_batches=10**4),
bm.TrainingDataMonitoring([cost, mu, sig, rmse],
every_n_batches=10),
be.Printing(every_n_batches=100, after_epoch=False),
]
)
loop.run()
|
