forked from shawntan/neural-transducers
/
lstm.py
126 lines (100 loc) · 4.15 KB
/
lstm.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
import theano
import theano.tensor as T
import numpy as np
import cPickle as pickle
from itertools import izip
from theano_toolkit import utils as U
from theano_toolkit import updates
from theano_toolkit.parameters import Parameters
def build(P, name, input_size, hidden_size):
name_init_hidden = "init_%s_hidden" % name
name_init_cell = "init_%s_cell" % name
P[name_init_hidden] = (0.1 * 2) * (np.random.rand(hidden_size) - 0.5)
P[name_init_cell] = (0.1 * 2) * (np.random.rand(hidden_size) - 0.5)
step = build_step(P, name, input_size, hidden_size)
def lstm_layer(X, row_transform=lambda x: x):
init_hidden = T.tanh(P[name_init_hidden])
init_cell = P[name_init_cell]
def _step(x, prev_cell, prev_hid):
if row_transform != None:
x = row_transform(x)
return step(x, prev_cell, prev_hid)
[cell, hidden], _ = theano.scan(
_step,
sequences=[X],
outputs_info=[init_cell, init_hidden],
)
return cell, hidden
return lstm_layer
def build_step(P, name, input_size, hidden_size):
name_W_input = "W_%s_input" % name
name_W_hidden = "W_%s_hidden" % name
name_W_cell = "W_%s_cell" % name
name_b = "b_%s" % name
P[name_W_input] = (0.1 * 2) * (np.random.rand(input_size, hidden_size * 4) - 0.5)
P[name_W_hidden] = (0.1 * 2) * (np.random.rand(hidden_size, hidden_size * 4) - 0.5)
P[name_W_cell] = (0.1 * 2) * (np.random.rand(hidden_size, hidden_size * 3) - 0.5)
bias_init = np.zeros((hidden_size * 4,), dtype=np.float32)
bias_init[1 * hidden_size:2 * hidden_size] = 3.
P[name_b] = bias_init
biases = P[name_b]
V_if = P[name_W_cell][:, 0 * hidden_size:2 * hidden_size]
V_o = P[name_W_cell][:, 2 * hidden_size:3 * hidden_size]
b_i = biases[0 * hidden_size:1 * hidden_size]
b_f = biases[1 * hidden_size:2 * hidden_size]
b_c = biases[2 * hidden_size:3 * hidden_size]
b_o = biases[3 * hidden_size:4 * hidden_size]
def step(x, prev_cell, prev_hidden):
transformed_x = T.dot(x, P[name_W_input])
x_i = transformed_x[0 * hidden_size:1 * hidden_size]
x_f = transformed_x[1 * hidden_size:2 * hidden_size]
x_c = transformed_x[2 * hidden_size:3 * hidden_size]
x_o = transformed_x[3 * hidden_size:4 * hidden_size]
transformed_hid = T.dot(prev_hidden, P[name_W_hidden])
h_i = transformed_hid[0 * hidden_size:1 * hidden_size]
h_f = transformed_hid[1 * hidden_size:2 * hidden_size]
h_c = transformed_hid[2 * hidden_size:3 * hidden_size]
h_o = transformed_hid[3 * hidden_size:4 * hidden_size]
transformed_cell = T.dot(prev_cell, V_if)
c_i = transformed_cell[0 * hidden_size:1 * hidden_size]
c_f = transformed_cell[1 * hidden_size:2 * hidden_size]
in_lin = x_i + h_i + b_i + c_i
forget_lin = x_f + h_f + b_f + c_f
cell_lin = x_c + h_c + b_c
in_gate = T.nnet.sigmoid(in_lin)
forget_gate = T.nnet.sigmoid(forget_lin)
cell_updates = T.tanh(cell_lin)
cell = forget_gate * prev_cell + in_gate * cell_updates
out_lin = x_o + h_o + b_o + T.dot(cell, V_o)
out_gate = T.nnet.sigmoid(out_lin)
hid = out_gate * T.tanh(cell)
return cell, hid
return step
if __name__ == "__main__":
P = Parameters()
X = T.ivector('X')
P.V = np.zeros((8, 8), dtype=np.int32)
X_rep = P.V[X]
P.W_output = np.zeros((15, 8), dtype=np.int32)
lstm_layer = build(P,
name="test",
input_size=8,
hidden_size=15
)
_, hidden = lstm_layer(X_rep)
output = T.nnet.softmax(T.dot(hidden, P.W_output))
delay = 5
label = X[:-delay]
predicted = output[delay:]
cost = -T.sum(T.log(predicted[T.arange(predicted.shape[0]), label]))
params = P.values()
gradients = T.grad(cost, wrt=params)
train = theano.function(
inputs=[X],
outputs=cost,
updates=[(p, p - 0.01 * g) for p, g in zip(params, gradients)],
)
while True:
print train(
np.random.randint(0, 8, size=20).astype(np.int32)
)