-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathmain_MDB.py
executable file
·187 lines (167 loc) · 7.03 KB
/
main_MDB.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
184
185
186
187
"""
-*- coding: utf-8 -*-
@Time : 2021/10/21 10:17
@Author : Wang Ziming
@Email : [email protected]
@File : main_TD.py
"""
import os
import sys
import scipy.io as io
import numpy as np
import torch
import os
from tqdm import tqdm
from model.util import setup_seed
from torch.utils.tensorboard import SummaryWriter
from model.ptn_loader import sPtn
from model.surrogate_act import SurrogateHeaviside
from model.snn import STCADenseLayer, ReadoutLayer, SNN
# num of neurons in the first and the last layer
num_in = 384
num_class = 10
dt = 3e-3
def getMDB(root_path, train_batch_size=1, val_batch_size=1, mode='spike'):
data = io.loadmat(root_path)
train_source = data['TrainData']
test_source = data['TestData']
train_data = sPtn(dataSource=train_source['ptn'][0, 0], labelSource=train_source['Labels'][0, 0].squeeze(),
TmaxSource=train_source['Tmax'][0, 0], mode='neuron',
dt=dt)
test_data = sPtn(dataSource=test_source['ptn'][0, 0], labelSource=test_source['Labels'][0, 0].squeeze(),
TmaxSource=test_source['Tmax'][0, 0],
mode='neuron',
dt=dt)
train_ldr = torch.utils.data.DataLoader(dataset=train_data, batch_size=train_batch_size, shuffle=True,
pin_memory=False, num_workers=0)
test_ldr = torch.utils.data.DataLoader(dataset=test_data, batch_size=val_batch_size, shuffle=False,
pin_memory=False, num_workers=0)
return train_ldr, test_ldr
def runTrain(train_ldr, optimizer, snn, evaluator):
global device, dtype
loss_record = []
predict_tot = []
label_tot = []
snn.train()
for idx, (ptns, labels) in enumerate(train_ldr):
ptns = ptns.permute([0, 2, 1]).contiguous().to(device, dtype)
# print(ptns.shape)
labels = labels.to(device)
# print(labels.dtype,labels.shape)
optimizer.zero_grad()
output, _ = snn(ptns)
# target = torch.nn.functional.one_hot(labels)
# print(target.shape)
loss = evaluator(output, labels)
loss.backward()
torch.nn.utils.clip_grad_value_(snn.parameters(), 5)
optimizer.step()
snn.clamp()
predict = torch.argmax(output, axis=1)
loss_record.append(loss.detach().cpu())
# print(torch.mean((predict==labels).float()))
predict_tot.append(predict)
# print(predict.shape)
label_tot.append(labels)
# print(torch.mean((predict == labels).float()))
predict_tot = torch.cat(predict_tot)
label_tot = torch.cat(label_tot)
train_acc = torch.mean((predict_tot == label_tot).float())
train_loss = torch.tensor(loss_record).sum() / len(label_tot)
return train_acc, train_loss
def runTest(val_ldr, snn, evaluator):
global test_trace, device, dtype
snn.eval()
with torch.no_grad():
loss_record = []
predict_tot = []
label_tot = []
for idx, (ptns, labels) in enumerate(val_ldr):
ptns = ptns.permute([0, 2, 1]).contiguous().to(device, dtype)
labels = labels.to(device)
output, _ = snn(ptns)
loss = evaluator(output, labels)
loss_record.append(loss)
# snn.clamp()
predict = torch.argmax(output, axis=1)
predict_tot.append(predict)
label_tot.append(labels)
predict_tot = torch.cat(predict_tot)
label_tot = torch.cat(label_tot)
val_acc = torch.mean((predict_tot == label_tot).float())
val_loss = torch.tensor(loss_record).sum() / len(label_tot)
return val_acc, val_loss
def main():
model_list = [name for name in os.listdir(save_dir) if name.startswith(prefix)]
file = prefix + 'cas_' + str(len(model_list))
save_path = os.path.join(save_dir, file)
train_ldr, val_ldr = getMDB(root_path='./data/M10_Data.mat', train_batch_size=64, val_batch_size=100)
print('Finish loading MedlyDB from: ', save_path)
# STCA full connected neural network
layers = []
# SurrogateHeaviside.sigma = 2
spike_fn = SurrogateHeaviside.apply
layers.append(STCADenseLayer(structure[0], structure[1], spike_fn, w_init_mean, w_init_std, recurrent=False,
lateral_connections=False, fc_drop=0.5))
# layers.append(SpikingDenseLayer(structure[1],structure[2],spike_fn,w_init_mean,w_init_std,recurrent=False,lateral_connections=True))
layers.append(ReadoutLayer(structure[1], structure[2], w_init_mean, w_init_std))
snn = SNN(layers).to(device, dtype)
# optimizer = RAdam(snn.parameters(), lr=3e-4)
# optimizer = torch.optim.AdamW(snn.parameters(), lr=3e-4, weight_decay=1e-3)
# optimizer = torch.optim.AdamW(snn.parameters(), lr=3e-4)
optimizer = torch.optim.Adam(snn.parameters(), lr=3e-4, amsgrad=False)
# optimizer = torch.optim.AdamW(snn.parameters(), lr=3e-4, amsgrad=True)
evaluator = torch.nn.CrossEntropyLoss()
# define some hyperparameter and holk for a run
if not os.path.exists(save_dir):
os.mkdir(save_dir)
best_acc = 0
test_trace = []
train_trace = []
loss_trace = []
if (not os.path.exists(log_dir)):
os.mkdir(log_dir)
start_epoch = 0
# resume
if (load_if):
state = torch.load(os.path.join(save_dir, file + '.t7'))
snn.load_state_dict(state['best_net'])
start_epoch = state['best_epoch']
train_trace = state['traces']['train']
test_trace = state['traces']['test']
loss_trace = state['traces']['loss']
# run forward - backward
for epoch in tqdm(range(start_epoch, start_epoch + epoch_num)):
train_acc, train_loss = runTrain(train_ldr, optimizer, snn, evaluator)
train_trace.append(train_acc)
loss_trace.append(train_loss)
print('\ntrain record: ', train_loss, train_acc)
val_acc, val_loss = runTest(val_ldr, snn, evaluator)
test_trace.append(val_acc)
print('validation record:', val_loss, val_acc)
if (val_acc > best_acc):
best_acc = val_acc
print('Saving model.. with acc {0} in the epoch {1}'.format(best_acc, epoch))
state = {
'best_acc': best_acc,
'best_epoch': epoch,
'best_net': snn.state_dict(),
'traces': {'train': train_trace, 'test': test_trace, 'loss': loss_trace},
# 'raw_pth': ptn_path
}
torch.save(state, os.path.join(save_path + '.t7'))
if __name__ == '__main__':
# global config
setup_seed(20)
device = torch.device('cuda:1' if torch.cuda.is_available() else 'cpu')
dtype = torch.float
structure = [num_in, 700, num_class]
w_init_std = 0.1
w_init_mean = 0.
epoch_num = 60
load_if = False
save_dir = './checkpoints'
log_dir = './log'
prefix = 'MDB10_SPG_'
print(device, dtype)
main()