-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathCNN+LSTM.py
137 lines (118 loc) · 5.9 KB
/
CNN+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
127
128
129
130
131
132
import torch.nn as nn
import torch.optim as optim
import torch
from torch.autograd import Variable
import matplotlib.pyplot as plt
import os
import numpy as np
from Utils import *
batch_size = 16
epochs = 50
c_input = Variable(torch.FloatTensor(1, 2, 200, 200)).cuda()
l_input = Variable(torch.FloatTensor(2, 2, 200, 200))
e_input = Variable(torch.FloatTensor(16, 9))
class Autodecoder(nn.Module):
def __init__(self):
super(Autodecoder, self).__init__()
self.conv1 = nn.Conv2d(in_channels=2, out_channels=4, kernel_size=5, stride=1, bias=True)
self.conv2 = nn.Conv2d(in_channels=4, out_channels=4, kernel_size=5, stride=2, bias=True)
self.conv3 = nn.Conv2d(in_channels=4, out_channels=8, kernel_size=5, stride=2, bias=True)
self.conv4 = nn.Conv2d(in_channels=8, out_channels=16, kernel_size=3, stride=2, bias=True)
self.conv5 = nn.Conv2d(in_channels=16, out_channels=16, kernel_size=3, stride=2, bias=True)
self.conv6 = nn.Conv2d(in_channels=16, out_channels=16, kernel_size=3, stride=2, bias=True)
self.conv7 = nn.Conv2d(in_channels=16, out_channels=16, kernel_size=3, stride=1, bias=True)
self.fc1 = nn.Linear(in_features=64, out_features=64)
self.relu = nn.ReLU()
self.deconv7 = nn.ConvTranspose2d(in_channels=16, out_channels=16, kernel_size=3, stride=1, bias=True)
self.deconv6 = nn.ConvTranspose2d(in_channels=16, out_channels=16, kernel_size=4, stride=2, bias=True)
self.deconv5 = nn.ConvTranspose2d(in_channels=16, out_channels=16, kernel_size=4, stride=2, bias=True)
self.deconv4 = nn.ConvTranspose2d(in_channels=16, out_channels=8, kernel_size=4, stride=2, bias=True)
self.deconv3 = nn.ConvTranspose2d(in_channels=8, out_channels=4, kernel_size=6, stride=2, bias=True)
self.deconv2 = nn.ConvTranspose2d(in_channels=4, out_channels=4, kernel_size=6, stride=2, bias=True)
self.deconv1 = nn.ConvTranspose2d(in_channels=4, out_channels=2, kernel_size=5, stride=1, bias=True)
def forward(self, input):
input = self.relu(self.conv1(input))
input = self.relu(self.conv2(input))
input = self.relu(self.conv3(input))
input = self.relu(self.conv4(input))
input = self.relu(self.conv5(input))
input = self.relu(self.conv6(input))
input = self.relu(self.conv7(input))
input = self.relu(self.fc1(input.view(batch_size, -1))).view(batch_size, -1, 2, 2)
input = self.relu(self.deconv7(input))
input = self.relu(self.deconv6(input))
input = self.relu(self.deconv5(input))
input = self.relu(self.deconv4(input))
input = self.relu(self.deconv3(input))
input = self.relu(self.deconv2(input))
input = self.deconv1(input)
return input
pa_path = os.path.abspath(os.path.join(os.path.dirname(__file__),os.path.pardir)) # get last state file name
name1 = '/SmartST/data/data(normalized)/'
name_start = 20161101
num_file = 30
path1 = pa_path + name1 + str(name_start)+'(normalized).npy'
temp_data = np.load(path1)
print("loading " + str(name_start) + "th file!")
for i in range(name_start+1, name_start+num_file):
print("loading "+str(i)+"th file!")
path = pa_path + name1 + str(i)+'(normalized).npy'
temp_data = np.concatenate((temp_data, np.load(path)), axis=0)
print("number of data is: "+str(np.shape(temp_data)[0]))
assert np.shape(temp_data)[0] == num_file*config.time, "Not reasonable data!!!"
intervals = config.intervals # [1 day,20 min,143(label)]
#### apply data_loader to pre-process data that fit to our network
# use for test
processed_data = data_loader(temp_data, intervals) # (4176,),c:(200, 200, 4),p:(200, 200, 4),l:(200, 200, 2)
train_data, test_data = create_train_test(processed_data)
train_size = train_data.__len__()
iter_size = int(train_size/batch_size)
train_loss = []
test_loss = []
model = Autodecoder().cuda()
Loss = nn.MSELoss()
optimizer = optim.Adam(model.parameters(), lr=1e-3, betas=(0.9, 0.999), eps=1e-8)
for epoch in range(epochs):
np.random.shuffle(train_data)
moving_loss = 0
counter = 0
for i in range(iter_size):
model.train(True)
batch_memory = np.array(train_data)[i * batch_size:(i + 1) * batch_size]
label = Variable(torch.from_numpy(np.array([memory_unit.label for memory_unit in batch_memory]))).view(-1, 2, 200, 200).float().cuda()
output = model(label)
optimizer.zero_grad()
loss = Loss(output, label)
loss.backward()
optimizer.step()
moving_loss += loss.data
counter += 1
if i % 10 == 0:
print('Training Epoch: {}, iteration: {}, loss: {}'.format(epoch, i, loss.data))
if i % 50 == 0:
moving_loss = 0
train_loss.append(moving_loss/counter)
counter = 0
with torch.no_grad():
model.train(False)
test_moving_loss = 0
torch.save(model.state_dict(), 'params_{}.pkl'.format(epoch))
test_iter = int(test_data.__len__()/batch_size)
for j in range(test_iter):
test_batch_memory = test_data[j * batch_size: (j+1) * batch_size]
test_label = Variable(torch.from_numpy(np.array([unit.label for unit in test_batch_memory]))).view(-1, 2, 200, 200).float().cuda()
test_out = model(test_label)
te_loss = Loss(test_out, test_label)
test_moving_loss += te_loss.data
test_loss.append(test_moving_loss/test_iter)
print('Test Epoch: {}, iteration: {}, test loss: {}'.format(epoch, i, te_loss.data))
plt.switch_backend('agg')
plt.figure('Learning Curve')
plt.xlabel('Iteration')
plt.ylabel('MSE Loss')
x_train = np.arange(len(train_loss))
p1 = plt.plot(x_train, train_loss, 'r')
p2 = plt.plot(x_train, test_loss, 'b')
plt.legend([p1[0], p2[0]], ['Train', 'Test'])
plt.savefig('train_test_result_epoch_{:d}'.format(epochs))
print('All Finished')