-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathmain.py
143 lines (118 loc) · 3.86 KB
/
main.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
import cv2
import numpy as np
import os
import torch
import torch.nn as nn
from torch.optim import Adam, SGD
from torch.autograd import Variable
import time
import torch.nn.functional as F
import matplotlib.pyplot as plt
import pickle
from Utils import process, getVal
from brain import CNNModel
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
dataV, labelV, input_size_val = getVal(device)
def train(model, optimizer):
model.train()
loss_sum =0
acc_sum = 0
input_size=0
optimizer.zero_grad()
for i in range(0,5):
file = open('./dataset/data'+str(i)+'.pkl', 'rb')
data= pickle.load(file)
file.close()
file = open('./dataset/label'+str(i)+'.pkl', 'rb')
label= pickle.load(file)
file.close()
input_size += len(data)
n,h,w = data.shape
data = data.reshape(n,1,h,w)
data = torch.from_numpy(data)
data = data.type(torch.FloatTensor)
label = label.astype(int)
label = torch.from_numpy(label)
data , label = data.to(device), label.to(device)
data, label = Variable(data), Variable(label)
for j in range(0,len(data),64):
bdata = data[j:j+64]
blabel= label[j:j+64]
output = model(bdata)
loss = F.cross_entropy(output, blabel)
loss_sum += loss.data.item()
loss.backward()
optimizer.step()
predict = output.data.max(1)[1]
acc = predict.eq(blabel.data).cpu().sum()
acc_sum +=acc
return loss_sum/input_size ,acc_sum.item()/input_size
def evaluate(model):
model.eval()
loss_sum =0
acc_sum = 0
for j in range(0,input_size_val,64):
bdata = dataV[j:j+64]
blabel= labelV[j:j+64]
output = model(bdata)
loss = F.cross_entropy(output, blabel)
loss_sum += loss.data.item()
predict = output.data.max(1)[1]
acc = predict.eq(blabel.data).cpu().sum()
acc_sum +=acc
return loss_sum/input_size_val ,acc_sum.item()/input_size_val
#initializing seed
torch.manual_seed(9372)
# np.random.seed(0)
#Model and optimizer
model = CNNModel()
optimizer = Adam(model.parameters(),0.0001)
#Cpu or Gpu training
model.to(device)
torch.save(model.state_dict(), "./weightedmodel.pth")
best_valid_loss = 1e5
change = 0
strikes =0
status = 'keep_train'
train_losses, val_losses =[],[]
train_accs, val_accs =[],[]
for epoch in range(200):
print('Epoch', epoch, status)
train_loss, train_acc = train(model, optimizer)
print('\t Train loss, accuracy', train_loss, train_acc)
valid_loss, valid_acc = evaluate(model)
print('\t Valid loss, best loss, accuracy', valid_loss, best_valid_loss, valid_acc )
train_losses.append(train_loss)
val_losses.append(valid_loss)
train_accs.append(train_acc)
val_accs.append(valid_acc)
if valid_loss>best_valid_loss:
strikes = strikes+1
if strikes>=8:
change += 1
strikes = 0
print('Current lr change', change)
if change >=8:
torch.save(model.state_dict(), "./weightedmodel.pth")
break
else:
model.load_state_dict(torch.load("./weightedmodel.pth"))
lr = 0.0001 * np.power(0.1, change)
for pg in optimizer.param_groups:
pg['lr'] = lr
else:
strikes =0
status = 'keep_train'
best_valid_loss = valid_loss
torch.save(model.state_dict(), "./weightedmodel.pth")
plt.plot(train_losses, label='Training Loss')
plt.plot(val_losses, label='Validation Loss')
plt.legend(frameon=True)
plt.show()
plt.clf()
plt.plot(train_accs, label='Training Acc')
plt.plot(val_accs, label='Validation Acc')
plt.legend(frameon=True)
plt.show()
val_loss, val_acc = evaluate(model)
print('Validation loss, accuracy', val_loss, val_acc)