-
Notifications
You must be signed in to change notification settings - Fork 7
/
Copy pathoptimizers.py
150 lines (117 loc) · 5.23 KB
/
optimizers.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
import theano
from theano import tensor
import numpy as np
import six
from utils import (
itemlist,
)
import settings
profile = settings.profile
# name(hyperp, tparams, grads, inputs (list), cost) = f_grad_shared, f_update
def adam(lr, tparams, grads, inp, cost):
gshared = [theano.shared(p.get_value() * 0.,
name='%s_grad' % k)
for k, p in six.iteritems(tparams)]
gsup = [(gs, g) for gs, g in zip(gshared, grads)]
f_grad_shared = theano.function(inp, cost, updates=gsup, profile=profile)
lr0 = 0.0002
b1 = 0.1
b2 = 0.001
e = 1e-8
updates = []
i = theano.shared(np.float32(0.))
i_t = i + 1.
fix1 = 1. - b1**(i_t)
fix2 = 1. - b2**(i_t)
lr_t = lr0 * (tensor.sqrt(fix2) / fix1)
for p, g in zip(tparams.values(), gshared):
m = theano.shared(p.get_value() * 0.)
v = theano.shared(p.get_value() * 0.)
m_t = (b1 * g) + ((1. - b1) * m)
v_t = (b2 * tensor.sqr(g)) + ((1. - b2) * v)
g_t = m_t / (tensor.sqrt(v_t) + e)
p_t = p - (lr_t * g_t)
updates.append((m, m_t))
updates.append((v, v_t))
updates.append((p, p_t))
updates.append((i, i_t))
f_update = theano.function([lr],
[],
updates=updates,
on_unused_input='ignore',
profile=profile)
return f_grad_shared, f_update
def adadelta(lr, tparams, grads, inp, cost):
zipped_grads = [theano.shared(p.get_value() * np.float32(0.),
name='%s_grad' % k)
for k, p in six.iteritems(tparams)]
running_up2 = [theano.shared(p.get_value() * np.float32(0.),
name='%s_rup2' % k)
for k, p in six.iteritems(tparams)]
running_grads2 = [theano.shared(p.get_value() * np.float32(0.),
name='%s_rgrad2' % k)
for k, p in six.iteritems(tparams)]
zgup = [(zg, g) for zg, g in zip(zipped_grads, grads)]
rg2up = [(rg2, 0.95 * rg2 + 0.05 * (g**2))
for rg2, g in zip(running_grads2, grads)]
f_grad_shared = theano.function(inp,
cost,
updates=zgup + rg2up,
profile=profile)
updir = [-tensor.sqrt(ru2 + 1e-6) / tensor.sqrt(rg2 + 1e-6) * zg
for zg, ru2, rg2 in zip(zipped_grads, running_up2, running_grads2)
]
ru2up = [(ru2, 0.95 * ru2 + 0.05 * (ud**2))
for ru2, ud in zip(running_up2, updir)]
param_up = [(p, p + ud) for p, ud in zip(itemlist(tparams), updir)]
f_update = theano.function([lr],
[],
updates=ru2up + param_up,
on_unused_input='ignore',
profile=profile)
return f_grad_shared, f_update
def rmsprop(lr, tparams, grads, inp, cost):
zipped_grads = [theano.shared(p.get_value() * np.float32(0.),
name='%s_grad' % k)
for k, p in six.iteritems(tparams)]
running_grads = [theano.shared(p.get_value() * np.float32(0.),
name='%s_rgrad' % k)
for k, p in six.iteritems(tparams)]
running_grads2 = [theano.shared(p.get_value() * np.float32(0.),
name='%s_rgrad2' % k)
for k, p in six.iteritems(tparams)]
zgup = [(zg, g) for zg, g in zip(zipped_grads, grads)]
rgup = [(rg, 0.95 * rg + 0.05 * g) for rg, g in zip(running_grads, grads)]
rg2up = [(rg2, 0.95 * rg2 + 0.05 * (g**2))
for rg2, g in zip(running_grads2, grads)]
f_grad_shared = theano.function(inp,
cost,
updates=zgup + rgup + rg2up,
profile=profile)
updir = [theano.shared(p.get_value() * np.float32(0.),
name='%s_updir' % k)
for k, p in six.iteritems(tparams)]
updir_new = [(ud, 0.9 * ud - 1e-4 * zg / tensor.sqrt(rg2 - rg**2 + 1e-4))
for ud, zg, rg, rg2 in zip(updir, zipped_grads, running_grads,
running_grads2)]
param_up = [(p, p + udn[1]) for p, udn in zip(
itemlist(tparams), updir_new)]
f_update = theano.function([lr],
[],
updates=updir_new + param_up,
on_unused_input='ignore',
profile=profile)
return f_grad_shared, f_update
def sgd(lr, tparams, grads, x, mask, y, cost):
gshared = [theano.shared(p.get_value() * 0.,
name='%s_grad' % k)
for k, p in six.iteritems(tparams)]
gsup = [(gs, g) for gs, g in zip(gshared, grads)]
f_grad_shared = theano.function(
[x, mask, y],
cost,
updates=gsup,
profile=profile)
pup = [(p, p - lr * g) for p, g in zip(itemlist(tparams), gshared)]
f_update = theano.function([lr], [], updates=pup, profile=profile)
return f_grad_shared, f_update