-
Notifications
You must be signed in to change notification settings - Fork 0
/
torch_model.py
183 lines (143 loc) · 6.59 KB
/
torch_model.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
175
176
177
178
179
180
181
182
183
import numpy as np
from torch import nn as nn
import torch
import os
import json
import matplotlib.pyplot as plt
import config
class Abstract_Model(nn.Module):
"""Abstract Model class."""
def __init__(self, save_path):
super(Abstract_Model, self).__init__()
if save_path is None:
save_path = os.getcwd()
if not os.path.exists(save_path):
os.makedirs(save_path)
self.save_path = save_path
self.config = None
def save(self, name = 'model', epoch=None):
save_path = self.save_path
if epoch is not None:
save_path = os.path.join(save_path, 'epoch', str(epoch).zfill(4))
if not os.path.exists(save_path):
os.makedirs(save_path)
model_path = os.path.join(save_path, name + '.pth')
torch.save(self.state_dict(), model_path)
self.save_config(save_path)
print("Model saved in path: %s" % model_path)
def load(self, name = 'model', epoch=None):
save_path = self.save_path
if epoch is not None:
save_path = os.path.join(save_path, 'epoch', str(epoch).zfill(4))
save_path = os.path.join(save_path, name + '.pth')
self.load_state_dict(torch.load(save_path))
print("Model restored from path: {:s}".format(save_path))
def save_config(self, save_path):
config_dict = self.config.__dict__
with open(os.path.join(save_path, 'config.json'), 'w') as f:
json.dump(config_dict, f)
with open(os.path.join(save_path, 'config.txt'), "w") as f:
for k, v in config_dict.items():
f.write(str(k) + ' >>> ' + str(v) + '\n\n')
class Simple_Model(Abstract_Model):
def __init__(self, opts, isize, osize):
super(Simple_Model, self).__init__(opts.save_path)
self.hidden_size = opts.rnn_size
self.batch_size = opts.batch_size
self.config = opts
self.i2h = nn.Linear(isize, opts.rnn_size)
self.h_b = torch.nn.Parameter(.01 * torch.rand(opts.rnn_size))
self.h_w = torch.nn.Parameter(.01 * torch.rand(opts.rnn_size, opts.rnn_size))
mask = np.ones((opts.rnn_size, opts.rnn_size)).astype(np.float32)
np.fill_diagonal(mask, 0)
mask = torch.from_numpy(mask)
h_mask = torch.nn.Parameter(mask, requires_grad=False)
self.h_mask = h_mask
self.h2o = torch.nn.Linear(opts.rnn_size, osize)
def forward(self, input, hidden):
i = self.i2h(input)
h_effective = torch.mul(self.h_w, self.h_mask)
h = torch.matmul(hidden, h_effective)
hidden = torch.relu(i + h + self.h_b)
out = self.h2o(hidden)
return hidden, out
def initialZeroState(self):
return torch.zeros(self.batch_size, self.hidden_size)
class Constrained_Model(Abstract_Model):
def __init__(self, opts, isize, osize):
super(Constrained_Model, self).__init__(opts.save_path)
self.hidden_size = opts.rnn_size
self.batch_size = opts.batch_size
self.config = opts
input_position_size = isize - 2
input_velocity_size = 2
hidden_attractor_size = opts.state_size
hidden_shift_size = opts.rnn_size - opts.state_size
self.position2attractor = nn.Linear(input_position_size, hidden_attractor_size)
self.velocity2shift = nn.Linear(input_velocity_size, hidden_shift_size)
self.h_w = torch.nn.Parameter(.01 * torch.rand(opts.rnn_size, opts.rnn_size))
self.h_b = torch.nn.Parameter(.01 * torch.rand(opts.rnn_size))
mask = np.ones((opts.rnn_size, opts.rnn_size)).astype(np.float32)
np.fill_diagonal(mask, 0)
mask[opts.state_size:, opts.state_size:] = 0
mask = torch.from_numpy(mask)
h_mask = torch.nn.Parameter(mask, requires_grad=False)
self.h_mask = h_mask
self.h2o = torch.nn.Linear(opts.rnn_size, osize)
def forward(self, input, hidden):
i_attractor = self.position2attractor(input[:,:-2])
i_shift = self.velocity2shift(input[:,-2:])
i = torch.cat((i_attractor, i_shift), dim=1)
h_effective = torch.mul(self.h_w, self.h_mask)
h = torch.matmul(hidden, h_effective)
hidden = torch.relu(i + h + self.h_b)
out = self.h2o(hidden)
return hidden, out
def initialZeroState(self):
return torch.zeros(self.batch_size, self.hidden_size)
class EI_Model(Abstract_Model):
def __init__(self, opts, isize, osize):
super(EI_Model, self).__init__(opts.save_path)
self.hidden_size = opts.rnn_size
self.batch_size = opts.batch_size
self.config = opts
self.i2h = torch.nn.Parameter(.01 * torch.rand(isize, opts.rnn_size))
target = 2
alpha = 2
nE = int(opts.rnn_size * opts.prop_ex)
nI = opts.rnn_size - nE
E = np.random.gamma(shape=alpha, scale=target / (nE * alpha), size=[nE, opts.rnn_size])
I = np.random.gamma(shape=alpha, scale=target / (nI * alpha), size=[nI, opts.rnn_size])
EI = np.concatenate([E, I], axis=0).astype(np.float32)
self.h_w = torch.nn.Parameter(torch.from_numpy(EI))
self.h_b = torch.nn.Parameter(.01 * torch.rand(opts.rnn_size))
ei_mask = np.eye(opts.rnn_size).astype(np.float32)
ei_mask[nE:] *= -1
self.ei_mask = torch.nn.Parameter(torch.from_numpy(ei_mask), requires_grad=False)
mask = np.ones((opts.rnn_size, opts.rnn_size)).astype(np.float32)
np.fill_diagonal(mask, 0)
mask = torch.from_numpy(mask)
h_mask = torch.nn.Parameter(mask, requires_grad=False)
self.h_mask = h_mask
self.h2o_w = torch.nn.Parameter(.01 * torch.rand(opts.rnn_size, osize))
self.h2o_b = torch.nn.Parameter(.01 * torch.rand(osize))
def forward(self, input, hidden):
i = torch.matmul(input, torch.abs(self.i2h))
_h_effective = torch.abs(torch.mul(self.h_w, self.h_mask))
h_effective = torch.matmul(self.ei_mask, _h_effective)
h = torch.matmul(hidden, h_effective)
hidden = torch.relu(i + h + self.h_b)
h2o_effective = torch.matmul(self.ei_mask, torch.abs(self.h2o_w))
out = torch.matmul(hidden, h2o_effective) + self.h2o_b
return hidden, out
def initialZeroState(self):
return torch.zeros(self.batch_size, self.hidden_size)
def load_config(save_path, epoch=None):
if epoch is not None:
save_path = os.path.join(save_path, 'epoch', str(epoch).zfill(4))
with open(os.path.join(save_path, 'config.json'), 'r') as f:
config_dict = json.load(f)
c = config.modelConfig()
for key, val in config_dict.items():
setattr(c, key, val)
return c