-
Notifications
You must be signed in to change notification settings - Fork 0
/
model.py
161 lines (132 loc) · 5.38 KB
/
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
import torch
import torch.nn as nn
import torch.nn.functional as F
import numpy as np
from torch.autograd import Variable
import pandas as pd
from sklearn.model_selection import train_test_split
from PIL import Image
import argparse
import sys, os, shutil, random
parser = argparse.ArgumentParser(description='PyTorch Shallow CNN')
parser.add_argument("--resume", type=str, default="", help="continue: 1, start over: 0")
args = parser.parse_args()
artists = np.load("labels.npy")
input_images = np.load("input_images.npz.npy")
input_images = input_images.transpose(0, 3, 1, 2)
#input_images = input_images
x_train, x_valid, y_train, y_valid = train_test_split(input_images, artists, test_size=0.2)
def save_checkpoint(state, is_best=False, filename='checkpoint.pth.tar'):
torch.save(state, filename)
if is_best:
shutil.copyfile(filename, 'model_best.pth.tar')
class shallowCNN(nn.Module):
def __init__(self):
#self.config = config
super(shallowCNN, self).__init__()
# [in, out, kernel_size, stride, padding]
self.bn0 = nn.BatchNorm2d(3)
self.max_pool0 = nn.MaxPool2d(2, 2)
self.conv1 = nn.Conv2d(3, 16, 3, 1, 1)
self.bn1 = nn.BatchNorm2d(16)
self.max_pool1 = nn.MaxPool2d(2, 2)
self.conv2 = nn.Conv2d(16, 32, 3, 1, 1)
self.bn2 = nn.BatchNorm2d(32)
self.max_pool2 = nn.MaxPool2d(2, 2)
self.conv3 = nn.Conv2d(32, 64, 3, 1, 1)
self.bn3 = nn.BatchNorm2d(64)
self.max_pool3 = nn.MaxPool2d(2, 2)
self.conv4 = nn.Conv2d(64, 128, 3, 1, 1)
self.bn4 = nn.BatchNorm2d(128)
self.max_pool4 = nn.MaxPool2d(2, 2)
self.conv5 = nn.Conv2d(128, 256, 3, 1, 1)
self.bn5 = nn.BatchNorm2d(256)
self.max_pool5 = nn.MaxPool2d(2, 2)
self.linear1 = nn.Linear(256 * 4 * 4, 2048)
self.linear2 = nn.Linear(2048, 346)
def forward(self, x):
x = self.max_pool0(self.bn0(x))
x = self.max_pool1(F.leaky_relu(self.bn1(self.conv1(x))))
x = self.max_pool2(F.leaky_relu(self.bn2(self.conv2(x))))
x = self.max_pool3(F.leaky_relu(self.bn3(self.conv3(x))))
x = self.max_pool4(F.leaky_relu(self.bn4(self.conv4(x))))
x = self.max_pool5(F.leaky_relu(self.bn5(self.conv5(x))))
#print(x.size(), x.size(1) * x.size(2) * x.size(3))
#exit()
x = x.view(-1, x.size(1) * x.size(2) * x.size(3))
x = F.leaky_relu(self.linear1(x))
x = F.dropout(x, p=0.5)
x = F.log_softmax(self.linear2(x))
return x
model = shallowCNN()
print(repr(model))
model.cuda()
EPOCHS = 100
LR = 1e-4
BATCH_SIZE = 16
START_EPOCH = 0
best_prec = 0
optimizer = torch.optim.Adam(model.parameters(), lr=LR, weight_decay=5e-3)
loss_fn = nn.NLLLoss()
def load_model(model, optimizer):
print("=> loading checkpoint '{}'".format(args.resume))
checkpoint = torch.load(args.resume)
START_EPOCH = checkpoint["epoch"] + 1
model.load_state_dict(checkpoint['state_dict'])
optimizer.load_state_dict(checkpoint['optimizer'])
print("=> loaded checkpoint '{}' (epoch {})"
.format(args.resume, checkpoint['epoch']))
return model, optimizer
if args.resume:
model, optimizer = load_model(model, optimizer)
for epoch in range(START_EPOCH, EPOCHS):
model.train()
index = np.random.permutation(len(x_train))
x_train = x_train[index]
y_train = y_train[index]
loss_list = []
correct_list = []
for i in range(len(x_train) // BATCH_SIZE + 1):
model.zero_grad()
start_ix = i * BATCH_SIZE
end_ix = (i + 1) * BATCH_SIZE
x_batch = x_train[start_ix:end_ix]
y_batch = y_train[start_ix:end_ix]
x_batch = Variable(torch.from_numpy(x_batch).float(), requires_grad=False).cuda()
y_batch = Variable(torch.LongTensor(y_batch), requires_grad=False).cuda()
logits = model(x_batch)
pred = logits.data.max(1, keepdim=True)[1]
correct = pred.eq(y_batch.data.view_as(pred)).cpu().sum()
loss = loss_fn(logits, y_batch)
loss_list.append(loss.data[0])
loss.backward()
optimizer.step()
sys.stdout.write(" "*80 + "\r")
sys.stdout.write("Epoch: %d, Step %d/%d, Precision: %.4f, loss: %.4f\r" %
(epoch, i, len(x_train) // BATCH_SIZE + 1, correct / BATCH_SIZE, np.mean(loss_list)))
model.eval()
correct = 0
loss_list = []
for i in range(len(x_valid) // BATCH_SIZE + 1):
start_ix = i * BATCH_SIZE
end_ix = (i + 1) * BATCH_SIZE
x_batch = x_valid[start_ix:end_ix]
y_batch = y_valid[start_ix:end_ix]
x_batch = Variable(torch.from_numpy(x_batch).float(), requires_grad=False).cuda()
y_batch = Variable(torch.LongTensor(y_batch), requires_grad=False).cuda()
logits = model(x_batch)
loss = loss_fn(logits, y_batch)
loss_list.append(loss.data[0])
pred = logits.data.max(1, keepdim=True)[1]
correct += pred.eq(y_batch.data.view_as(pred)).cpu().sum()
correct /= len(x_valid)
sys.stdout.write(" "*80 + "\r")
print("Epoch: %d, precision: %.4f, loss: %.4f" % (epoch, correct, np.mean(loss_list)))
if correct > best_prec:
best_prec = correct
save_checkpoint({
'epoch': epoch + 1,
'state_dict': model.state_dict(),
'optimizer' : optimizer.state_dict(),
'best_prec': best_prec
}, best_prec == correct)