版权声明：本文为博主原创文章，欢迎转载，并请注明出处。联系方式：460356155@qq.com
在前一篇文章MINIST深度学习识别：python全连接神经网络和pytorch LeNet CNN网络训练实现及比较（一）中，直接用python创建全连接神经网络模型进行深度学习训练，这样可以对神经网络有较为深刻的认识。
但是在实际应用中，一般都是采用各种深度学习框架来开展人工智能项目，以下就采用pytorch来实现前一篇文章中的全连接神经网络（784-300-10）。
1 1 # -*- coding:utf-8 -*- 2 2 3 3 u"""pytorch LineNet神经网络训练学习MINIST""" 4 4 5 5 __author__ = 'zhengbiqing 460356155@qq.com' 6 6 7 7 8 8 import torch as t 9 9 import torchvision as tv 10 10 import torch.nn as nn 11 11 import torch.nn.functional as F 12 12 import torchvision.transforms as transforms 13 13 from torch.autograd import Variable 14 14 import matplotlib.pyplot as plt 15 15 import datetime 16 16 17 17 18 18 #是否训练网络 19 19 TRAIN = True 20 20 21 21 #是否保存网络 22 22 SAVE_PARA = False 23 23 24 24 #学习率和训练次数 25 25 LR = 0.05 26 26 EPOCH = 10 27 27 28 28 #训练每批次的样本数 29 29 BATCH_SZ = 16 30 30 31 31 #样本读取线程数 32 32 WORKERS = 4 33 33 34 34 #网络参赛保存文件名 35 35 PARAS_FN = 'minist_linenet_params.pkl' 36 36 37 37 #minist数据存放位置 38 38 ROOT = '/home/zbq/pytorch/minist' 39 39 40 40 41 41 #定义网络模型 42 42 class LineNet(nn.Module): 43 43 def __init__(self): 44 44 super(LineNet, self).__init__() 45 45 46 46 self.fc = nn.Sequential( 47 47 nn.Linear(28*28, 300), 48 48 nn.ReLU(), 49 49 nn.Linear(300, 10) 50 50 ) 51 51 52 52 def forward(self, x): 53 53 #x是2维tensor，转换为1维向量 54 54 x = x.view(x.size()[0], -1) 55 55 x = self.fc(x) 56 56 return x 57 57 58 58 59 59 ''' 60 60 训练并测试网络 61 61 net：网络模型 62 62 train_data_load：训练数据集 63 63 test_data_load：测试数据集 64 64 epochs：训练迭代次数 65 65 save：是否保存训练结果 66 66 ''' 67 67 def net_train(net, train_data_load, test_data_load, epochs, save): 68 68 start_time = datetime.datetime.now() 69 69 70 70 loss_list = [] 71 71 72 72 for epoch in range(epochs): 73 73 for i, data in enumerate(train_data_load, 0): 74 74 img, label = data 75 75 img, label = Variable(img), Variable(label) 76 76 img, label = img.cuda(), label.cuda() 77 77 78 78 optimizer.zero_grad() 79 79 80 80 pre = net(img) 81 81 loss = loss_func(pre, label) 82 82 loss.backward() 83 83 84 84 optimizer.step() 85 85 86 86 #显示损失函数值的变化 87 87 loss_data = loss.data.item() 88 88 if i % 1000 == 999: 89 89 print('epoch:{epoch} i:{i} loss:{loss}'.format(epoch=epoch, i=i, loss=loss_data)) 90 90 91 91 if i % 100 == 99: 92 92 loss_list.append(loss_data) 93 93 94 94 # 每个epoch结束后用测试集检查识别准确度 95 95 net_test(epoch, test_data_load) 96 96 97 97 print('MINIST pytorch LineNet Train: EPOCH:{epochs}, BATCH_SZ:{batch_sz}, LR:{lr}'.format(epochs=epochs, batch_sz=BATCH_SZ, lr=LR)) 98 98 print('train spend time: ', datetime.datetime.now() - start_time) 99 99 100100 if save: 101101 t.save(net.state_dict(), PARAS_FN) 102102 103103 #显示目标函数值的变化曲线 104104 plt.plot(loss_list) 105105 plt.show() 106106 107107 108108 ''' 109109 用测试集检查准确率 110110 ''' 111111 def net_test(epoch, test_data_load): 112112 ok = 0 113113 114114 for i, data in enumerate(test_data_load): 115115 img, label = data 116116 img, label = Variable(img), Variable(label) 117117 img, label = img.cuda(), label.cuda() 118118 119119 outs = net(img) 120120 _, pre = t.max(outs.data, 1) 121121 ok += (pre == label).sum() 122122 123123 acc = ok.item() * 100 / (len(test_data_load) * BATCH_SZ) 124124 125125 print('EPOCH:{epoch}, ACC:{acc}\n'.format(epoch=epoch, acc=acc)) 126126 127127 128128 #图像数值转换，ToTensor源码注释 129129 """Convert a ``PIL Image`` or ``numpy.ndarray`` to tensor. 130130 Converts a PIL Image or numpy.ndarray (H x W x C) in the range 131131 [0, 255] to a torch.FloatTensor of shape (C x H x W) in the range [0.0, 1.0]. 132132 """ 133133 #归一化，把[0.0, 1.0]变换为[-1,1], ([0, 1] - 0.5) / 0.5 = [-1, 1] 134134 transform = tv.transforms.Compose([transforms.ToTensor(), transforms.Normalize([0.5], [0.5])]) 135135 136136 #定义数据集 137137 train_data = tv.datasets.MNIST(root=ROOT, train=True, download=True, transform=transform) 138138 test_data = tv.datasets.MNIST(root=ROOT, train=False, download=False, transform=transform) 139139 140140 train_load = t.utils.data.DataLoader(train_data, batch_size=BATCH_SZ, shuffle=True, num_workers=WORKERS) 141141 test_load = t.utils.data.DataLoader(test_data, batch_size=BATCH_SZ, shuffle=False, num_workers=WORKERS) 142142 143143 print('train data num:', len(train_data), ', test data num:', len(test_data)) 144144 145145 146146 net = LineNet() 147147 net.cuda() 148148 149149 loss_func = nn.CrossEntropyLoss() 150150 optimizer = t.optim.SGD(net.parameters(), lr=LR) 151151 152152 if TRAIN: 153153 net_train(net, train_load, test_load, EPOCH, SAVE_PARA) 154154 else: 155155 net.load_state_dict(t.load(PARAS_FN)) 156156 net_test(0, test_load)
网络训练结果准确率基本在97%～98%，和前一篇MINIST深度学习识别：python全连接神经网络和pytorch LeNet CNN网络训练实现及比较（一）相同网络结构的全连接神经网络相当，但是因为这里采用GPU运算，训练时间降低到1/8。
此外，借助pytorch，代码更简单。
运行结果如下：
train data num: 60000 , test data num: 10000
epoch:0 i:999 loss:0.3457891643047333
epoch:0 i:1999 loss:0.09639787673950195
epoch:0 i:2999 loss:0.27898865938186646
EPOCH:0, ACC:94.84
epoch:1 i:999 loss:0.33745211362838745
epoch:1 i:1999 loss:0.11106520891189575
epoch:1 i:2999 loss:0.21725007891654968
EPOCH:1, ACC:96.42
epoch:2 i:999 loss:0.3825737535953522
epoch:2 i:1999 loss:0.02866300940513611
epoch:2 i:2999 loss:0.11832481622695923
EPOCH:2, ACC:96.77
epoch:3 i:999 loss:0.11886310577392578
epoch:3 i:1999 loss:0.012149035930633545
epoch:3 i:2999 loss:0.030409961938858032
EPOCH:3, ACC:97.2
epoch:4 i:999 loss:0.008915185928344727
epoch:4 i:1999 loss:0.008089780807495117
epoch:4 i:2999 loss:0.0005310177803039551
EPOCH:4, ACC:97.6
epoch:5 i:999 loss:0.02993696928024292
epoch:5 i:1999 loss:0.01784616708755493
epoch:5 i:2999 loss:0.10544028878211975
EPOCH:5, ACC:97.6
epoch:6 i:999 loss:0.008486062288284302
epoch:6 i:1999 loss:0.0334945023059845
epoch:6 i:2999 loss:0.00291365385055542
EPOCH:6, ACC:97.37
epoch:7 i:999 loss:0.0062919557094573975
epoch:7 i:1999 loss:0.0003241896629333496
epoch:7 i:2999 loss:0.0006818175315856934
EPOCH:7, ACC:97.23
epoch:8 i:999 loss:0.0007421970367431641
epoch:8 i:1999 loss:0.005641639232635498
epoch:8 i:2999 loss:0.005949795246124268
EPOCH:8, ACC:97.7
epoch:9 i:999 loss:0.024028539657592773
epoch:9 i:1999 loss:0.005388796329498291
epoch:9 i:2999 loss:0.0029097795486450195
EPOCH:9, ACC:97.39
MINIST pytorch LineNet Train: EPOCH:10, BATCH_SZ:16, LR:0.05
train spend time: 0:00:43.183836
损失函数值变化曲线为：
