forked from nd-hung/DL4DistancePrediction2
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathAdams.py
175 lines (136 loc) · 5.55 KB
/
Adams.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
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
"""
The MIT License (MIT)
Copyright (c) 2015 Alec Radford
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.
"""
import theano
import theano.tensor as T
import numpy as np
def Adam(params, grads, lr=0.0002, b1=0.1, b2=0.001, e=1e-8):
updates = []
other_params = []
#grads = T.grad(cost, params)
i = theano.shared(np.float32(0.).astype(theano.config.floatX))
other_params.append(i)
i_t = i + 1.
fix1 = 1. - (1. - b1)**i_t
fix2 = 1. - (1. - b2)**i_t
lr_t = lr * (T.sqrt(fix2) / fix1)
for p, g in zip(params, grads):
m = theano.shared(p.get_value() * 0.)
v = theano.shared(p.get_value() * 0.)
other_params.append(m)
other_params.append(v)
m_t = (b1 * g) + ((1. - b1) * m)
v_t = (b2 * T.sqr(g)) + ((1. - b2) * v)
g_t = m_t / (T.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))
return updates, other_params
def AMSGrad(params, grads, lr=0.0002, b1=0.1, b2=0.001, e=1e-8):
updates = []
other_params = []
#grads = T.grad(cost, params)
i = theano.shared(np.float32(0.).astype(theano.config.floatX))
other_params.append(i)
i_t = i + 1.
fix1 = 1. - (1. - b1)**i_t
fix2 = 1. - (1. - b2)**i_t
lr_t = lr * (T.sqrt(fix2) / fix1)
for p, g in zip(params, grads):
m = theano.shared(p.get_value() * 0.)
v = theano.shared(p.get_value() * 0.)
v_hat = theano.shared(p.get_value() * 0.)
other_params.append(m)
other_params.append(v)
other_params.append(v_hat)
m_t = (b1 * g) + ((1. - b1) * m)
v_t = (b2 * T.sqr(g)) + ((1. - b2) * v)
v_hat_t = T.maximum( v_hat, v_t)
g_t = m_t / (T.sqrt(v_hat_t) + e)
p_t = p - (lr_t * g_t)
updates.append((m, m_t))
updates.append((v, v_t))
updates.append((v_hat, v_hat_t) )
updates.append((p, p_t))
updates.append((i, i_t))
return updates, other_params
## pdecay has the same length of params. Each element in pdecay is either 0 or corresponds to params
def AdamW(params, grads, pdecay=None, l2reg=0.1, lr=0.0002, b1=0.1, b2=0.001, e=1e-8):
if pdecay is None:
wdecay = params
else:
wdecay = pdecay
updates = []
other_params = []
#grads = T.grad(cost, params)
i = theano.shared(np.float32(0.).astype(theano.config.floatX))
other_params.append(i)
i_t = i + 1.
fix1 = 1. - (1. - b1)**i_t
fix2 = 1. - (1. - b2)**i_t
lr_t = lr * (T.sqrt(fix2) / fix1)
for p, g, d in zip(params, grads, wdecay):
m = theano.shared(value=np.zeros(p.shape.eval(),dtype=theano.config.floatX), borrow=True)
v = theano.shared(value=np.zeros(p.shape.eval(),dtype=theano.config.floatX), borrow=True)
other_params.append(m)
other_params.append(v)
m_t = (b1 * g) + ((1. - b1) * m)
v_t = (b2 * T.sqr(g)) + ((1. - b2) * v)
g_t = m_t / (T.sqrt(v_t) + e)
p_t = p - lr_t * g_t - lr * l2reg * d
updates.append((m, m_t))
updates.append((v, v_t))
updates.append((p, p_t))
updates.append((i, i_t))
return updates, other_params
## pdecay has the same length of params. Each element in pdecay is either 0 or correspond to params
def AdamWAMS(params, grads, pdecay=None, l2reg=0.1, lr=0.0002, b1=0.1, b2=0.001, e=1e-8):
if pdecay is None:
wdecay = params
else:
wdecay = pdecay
updates = []
other_params = []
#grads = T.grad(cost, params)
i = theano.shared(np.float32(0.).astype(theano.config.floatX))
other_params.append(i)
i_t = i + 1.
fix1 = 1. - (1. - b1)**i_t
fix2 = 1. - (1. - b2)**i_t
lr_t = lr * (T.sqrt(fix2) / fix1)
for p, g, d in zip(params, grads, wdecay):
m = theano.shared(value=np.zeros(p.shape.eval(),dtype=theano.config.floatX), borrow=True)
v = theano.shared(value=np.zeros(p.shape.eval(),dtype=theano.config.floatX), borrow=True)
v_hat = theano.shared(value=np.zeros(p.shape.eval(),dtype=theano.config.floatX), borrow=True)
other_params.append(m)
other_params.append(v)
other_params.append(v_hat)
m_t = (b1 * g) + ((1. - b1) * m)
v_t = (b2 * T.sqr(g)) + ((1. - b2) * v)
v_hat_t = T.maximum( v_hat, v_t)
g_t = m_t / (T.sqrt(v_hat_t) + e)
p_t = p - lr_t * g_t - lr * l2reg * d
updates.append((m, m_t))
updates.append((v, v_t))
updates.append((v_hat, v_hat_t) )
updates.append((p, p_t))
updates.append((i, i_t))
return updates, other_params