stack-autoencoder/trian.py at master · zhangxu0307/stack-autoencoder · GitHub

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
import torch as th
import torchvision
from torch.autograd import Variable
from torch import nn
from torch import optim
from torchvision import datasets
import torchvision.transforms as transforms
from SAE import *


def loadMNIST(batchSize):

    root = "./data/"

    trans= transforms.Compose([transforms.ToTensor(), transforms.Normalize((0.1307,), (0.3081,))])

    train_set = datasets.MNIST(root=root, train=True, transform=trans, download=True)
    test_set = datasets.MNIST(root=root, train=False, transform=trans)

    train_loader = th.utils.data.DataLoader(dataset=train_set, batch_size=batchSize, shuffle=True)
    test_loader = th.utils.data.DataLoader(dataset=test_set, batch_size=batchSize, shuffle=False)

    print ('==>>> total trainning batch number: {}'.format(len(train_loader)))
    print ('==>>> total testing batch number: {}'.format(len(test_loader)))

    return train_loader, test_loader


def trainAE(encoderList, trainLayer, batchSize, epoch, useCuda = False):

    if useCuda:
        for i in range(len(encoderList)):
            encoderList[i].cuda()

    optimizer = optim.SGD(encoderList[trainLayer].parameters(), lr=0.1)
    ceriation = nn.L1Loss()
    trainLoader, testLoader = loadMNIST(batchSize=batchSize)

    for i in range(epoch):

        sum_loss = 0

        if trainLayer != 0: # 单独处理第0层，因为第一个编码器之前没有前驱的编码器了
            for i in range(trainLayer): # 冻结要训练前面的所有参数
                for param in encoderList[i].parameters():
                    param.requires_grad = False

        for batch_idx, (x, target) in enumerate(trainLoader):
            optimizer.zero_grad()
            if useCuda:
                x, target = x.cuda(), target.cuda()
            x, target = Variable(x), Variable(target)
            x = x.view(-1, 784)
            # 产生需要训练层的输入数据
            out = x
            if trainLayer != 0:
                for i in range(trainLayer):
                    out = encoderList[i](out, rep=True)

            # 训练指定的自编码器
            pred = encoderList[trainLayer](out, rep=False)

            loss = ceriation(pred, out)
            sum_loss += loss.data[0]
            loss.backward()
            optimizer.step()

            if (batch_idx + 1) % 10 == 0 or (batch_idx + 1) == len(trainLoader):
                print('==>>> train layer:{}, epoch: {}, batch index: {}, train loss: {:.6f}'
                      .format(trainLayer, i, batch_idx + 1, sum_loss/batch_idx))


def trainClassifier(model, batchSize, epoch, useCuda = False):

    if useCuda:
        model = model.cuda()

    # 解锁参数
    for param in model.parameters():
        param.requires_grad = True

    optimizer = optim.SGD(model.parameters(), lr=0.1)
    ceriation = nn.NLLLoss()
    trainLoader, testLoader = loadMNIST(batchSize=batchSize)

    for i in range(epoch):

        # trainning
        sum_loss = 0

        for batch_idx, (x, target) in enumerate(trainLoader):
            optimizer.zero_grad()
            if useCuda:
                x, target = x.cuda(), target.cuda()
            x, target = Variable(x), Variable(target)
            x = x.view(-1, 784)

            out = model(x)

            loss = ceriation(out, target)
            sum_loss += loss.data[0]
            loss.backward()
            optimizer.step()

            if (batch_idx + 1) % 10 == 0 or (batch_idx + 1) == len(trainLoader):
                print('==>>> epoch: {}, batch index: {}, train loss: {:.6f}'.format( i, batch_idx + 1, sum_loss/batch_idx))

        # testing
        correct_cnt, sum_loss = 0, 0
        total_cnt = 0
        for batch_idx, (x, target) in enumerate(testLoader):

            x, target = Variable(x, volatile=True), Variable(target, volatile=True)
            if useCuda:
                x, target = x.cuda(), target.cuda()
            x = x.view(-1, 784)

            out = model(x)
            loss = ceriation(out, target)
            _, pred_label = th.max(out.data, 1)
            total_cnt += x.data.size()[0]
            correct_cnt += (pred_label == target.data).sum()

            # smooth average
            if (batch_idx + 1) % 100 == 0 or (batch_idx + 1) == len(testLoader):
                print('==>>> epoch: {}, batch index: {}, test loss: {:.6f}, acc: {:.3f}'.format(
                    i, batch_idx + 1, sum_loss/batch_idx, correct_cnt * 1.0 / total_cnt))

if __name__ == '__main__':

    batchSize = 128
    AEepoch = 20
    epoch = 10

    encoder1 = AutoEncoder(784, 256)
    encoder2 = AutoEncoder(256, 64)
    #encoder3 = AutoEncoder(196, 98)

    encoderList = [encoder1, encoder2,]

    trainAE(encoderList, 0, batchSize, AEepoch, useCuda=True)
    trainAE(encoderList, 1, batchSize, AEepoch, useCuda=True)
    #trainAE(encoderList, 2, batchSize, AEepoch, useCuda=True)

    model = SAE(encoderList)
    #model = MLP()
    trainClassifier(model, batchSize, epoch, useCuda=True)