import torch
from torch.utils.data import Dataset
from torch.utils.data import DataLoader
import torch.nn.functional as F
from torch import nn
import numpy as np
import scipy.ndimage
import time
import math
import csv
import os
import matplotlib.pyplot as plt
batch_size = 64
class DiabetesDataset(Dataset):
def __init__(self, filepath,UseRotation = False):
xy = np.loadtxt(filepath, delimiter=',', dtype=np.float32)
#self.len = xy.shape[0]
self.x_data = torch.from_numpy(xy[:, 1:])
if UseRotation:
#加上左右旋转10度的图片数据
self.x_data_plus10 = self.x_data
for i, x in enumerate(self.x_data_plus10, 0):
x1 = scipy.ndimage.interpolation.rotate(x.reshape(28, 28), 10, cval=0.0,
reshape=False).reshape(784)
self.x_data_plus10[i] = torch.from_numpy(x1)
self.x_data_minus10 = self.x_data
for i, x in enumerate(self.x_data_minus10, 0):
x1 = scipy.ndimage.interpolation.rotate(x.reshape(28, 28), -10, cval=0.0,
reshape=False).reshape(784)
self.x_data_minus10[i] = torch.from_numpy(x1)
self.x_data = torch.cat([self.x_data, self.x_data_plus10, self.x_data_minus10], dim=0)
self.x_data = (self.x_data/255 - 0.1307)/0.3081 #映射到（0，1）分布，很多时候可以加快收敛速度
print("x_data.shape = ",self.x_data.shape)
self.x_data = self.x_data.reshape(-1,1,28,28) #如果使用CNN需要加上这一句，DNN则注释掉
self.y_data = torch.from_numpy(xy[:, [0]])
self.y_data = self.y_data.squeeze(1).long()
if UseRotation:
self.y_data = torch.tile(self.y_data, [3])
print("y_data.shape = ",self.y_data.shape)
def __getitem__(self, index):
return self.x_data[index], self.y_data[index]
def __len__(self):
return len(self.x_data)
train_dataset = DiabetesDataset('mnist_train.csv',UseRotation=True)
train_loader = DataLoader(dataset=train_dataset, batch_size=batch_size, shuffle=True)
test_dataset = DiabetesDataset('mnist_test.csv')
test_loader = DataLoader(dataset=test_dataset, batch_size=batch_size, shuffle=False)
class InceptionA(torch.nn.Module):
def __init__(self, in_channels):
super(InceptionA, self).__init__()
self.branch1x1 = nn.Conv2d(in_channels, 16, kernel_size=1)
self.branch5x5_1 = nn.Conv2d(in_channels, 16, kernel_size=1)
self.branch5x5_2 = nn.Conv2d(16, 24, kernel_size=5, padding=2)
self.branch3x3_1 = nn.Conv2d(in_channels, 16, kernel_size=1)
self.branch3x3_2 = nn.Conv2d(16, 24, kernel_size=3, padding=1)
self.branch3x3_3 = nn.Conv2d(24, 24, kernel_size=3, padding=1)
self.branch_pool = nn.Conv2d(in_channels, 24, kernel_size=1)
def forward(self, x):
branch1x1 = self.branch1x1(x)
branch5x5 = self.branch5x5_1(x)
branch5x5 = self.branch5x5_2(branch5x5)
branch3x3 = self.branch3x3_1(x)
branch3x3 = self.branch3x3_2(branch3x3)
branch3x3 = self.branch3x3_3(branch3x3)
branch_pool = F.avg_pool2d(x, kernel_size=3, stride=1, padding=1)
branch_pool = self.branch_pool(branch_pool)
outputs = [branch1x1, branch5x5, branch3x3, branch_pool]
return torch.cat(outputs, dim=1)
class ResidualBlock(nn.Module):
def __init__(self, channels):
super(ResidualBlock, self).__init__()
self.channels = channels
self.conv1 = torch.nn.Conv2d(channels, channels, kernel_size=3, padding=1)
self.conv2 = torch.nn.Conv2d(channels, channels, kernel_size=3, padding=1)
def forward(self, x):
y = F.relu(self.conv1(x))
y = self.conv2(y)
return F.relu(x + y)
#全连接网络
class Net(torch.nn.Module):
def __init__(self):
super(Net, self).__init__()
self.l1 = torch.nn.Linear(784, 512)
self.l2 = torch.nn.Linear(512, 256)
self.l3 = torch.nn.Linear(256, 128)
self.l4 = torch.nn.Linear(128, 64)
self.l5 = torch.nn.Linear(64, 10)
self.sigmoid = torch.nn.Sigmoid()
def forward(self, x):
x = x.view(-1, 784)
x = F.relu(self.l1(x))
x = F.relu(self.l2(x))
x = F.relu(self.l3(x))
x = F.relu(self.l4(x))
return self.l5(x)
#卷积神经网络
class Net2(torch.nn.Module):
def __init__(self):
super(Net2, self).__init__()
self.conv1 = torch.nn.Conv2d(1, 10,kernel_size = 5)
self.conv2 = torch.nn.Conv2d(10, 20,kernel_size = 5)
self.pooling = torch.nn.MaxPool2d(2)
self.l1 = torch.nn.Linear(320, 256)
self.l2 = torch.nn.Linear(256, 128)
self.l3 = torch.nn.Linear(128, 64)
self.l4 = torch.nn.Linear(64, 10)
def forward(self, x):
#这里拿出第1维的个数即(n,1,28,28)中的第一个元素，是样本一个batch的样本个数
#注意输入数据是四维的
batch_size = x.size(0)
x = F.relu(self.pooling(self.conv1(x)))
x = F.relu(self.pooling(self.conv2(x)))
#把它变成全连接网络batch * 320
x = x.view(batch_size,-1)
x = F.relu(self.l1(x))
x = F.relu(self.l2(x))
x = F.relu(self.l3(x))
x = self.l4(x)
return x
#inception网络
class Net3(torch.nn.Module):
def __init__(self):
super(Net3, self).__init__()
self.conv1 = torch.nn.Conv2d(1, 10, kernel_size=5)
self.conv2 = torch.nn.Conv2d(88, 20, kernel_size=5)
self.incep1 = InceptionA(in_channels=10)
self.incep2 = InceptionA(in_channels=20)
self.mp = nn.MaxPool2d(2)
self.fc = torch.nn.Linear(1408, 10)
self.l1 = torch.nn.Linear(1408, 512)
self.l2 = torch.nn.Linear(512, 128)
self.l3 = torch.nn.Linear(128, 64)
self.l4 = torch.nn.Linear(64, 10)
def forward(self, x):
in_size = x.size(0)
# 变成10*12*12
x = F.relu(self.mp(self.conv1(x)))
# 变成88*12*12
x = self.incep1(x)
# 变成20*4*4
x = F.relu(self.mp(self.conv2(x)))
# 变成88*4*4
x = self.incep2(x)
# 展开成全连接层
x = x.view(in_size, -1)
x = F.relu(self.l1(x))
x = F.relu(self.l2(x))
x = F.relu(self.l3(x))
x = self.l4(x)
return x
#深度残差网络（ResNet）(四个网络中效果最好)
class Net4(torch.nn.Module):
def __init__(self):
super(Net4, self).__init__()
self.conv1 = torch.nn.Conv2d(1, 16, kernel_size=5)
self.conv2 = torch.nn.Conv2d(16, 32, kernel_size=5)
self.rblock1 = ResidualBlock(16)
self.rblock2 = ResidualBlock(32)
self.mp = nn.MaxPool2d(2)
self.l1 = torch.nn.Linear(512, 10)
#self.l2 = torch.nn.Linear(256, 128)
#self.l3 = torch.nn.Linear(128, 64)
#self.l4 = torch.nn.Linear(64, 10)
def forward(self, x):
in_size = x.size(0)
# 变成16*12*12
x = F.relu(self.mp(self.conv1(x)))
# 变成16*12*12
x = self.rblock1(x)
# 变成32*4*4
x = F.relu(self.mp(self.conv2(x)))
# 变成32*4*4
x = self.rblock2(x)
# 展开成全连接层
x = x.view(in_size, -1)
#x = F.relu(self.l1(x))
#x = F.relu(self.l2(x))
#x = F.relu(self.l3(x))
x = self.l1(x)
return x
model = Net4()
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
model.to(device)
criterion = torch.nn.CrossEntropyLoss()
#optimizer = torch.optim.SGD(model.parameters(),lr = 0.005,momentum=0.8,nesterov=True)
optimizer = torch.optim.Adam(model.parameters(),lr = 0.00025)
def time_since(since):
s = time.time() - since
m = math.floor(s / 60)
s -= m * 60
return '%dm %ds' % (m, s)
def train(epoch,train_loader):
running_loss = 0.0
for batch_idx, data in enumerate(train_loader, 0):
inputs, target = data
inputs, target = inputs.to(device), target.to(device)
optimizer.zero_grad()
outputs = model(inputs)
loss = criterion(outputs, target)
loss.backward()
optimizer.step()
running_loss += loss.item()
if batch_idx % 150 == 149:
print(f'[{
time_since(start)}] ', end='')
print('[%d,%5d] loss: %.3f' % (epoch + 1, batch_idx + 1, running_loss / 150))
running_loss = 0
def test(test_loader):
correct = 0
total = 0
pred = []
with torch.no_grad(): # 这部分代码不会计算梯度
for data in test_loader:
images, labels = data
images,labels = images.to(device),labels.to(device)
outputs = model(images)
# 这里用max函数找输出节点中的最大值（即输出矩阵中每一行的最大值），返回该值和对应下标
_, predicted = torch.max(outputs.data, dim=1)
#转换成列表，并入总的预测列表中
pred = pred + predicted.cpu().tolist()
# labels.size(0)返回行数，也即是样本个数
total += labels.size(0)
# 把两个N*1的Tensor做比较相等是1否则是0，把所有结果相加就是正确的个数
correct += (predicted == labels).sum().item()
print("Total:",total)
print("Corrcet:",correct)
print('Accuracy on test set: %.2f %%' % (100 * correct / total))
return pred,100 * correct / total
# 保存预测数据
def save_pred(preds, file):
''' Save predictions to specified file '''
print('Saving results to {}'.format(file))
with open(file, 'w', newline='') as fp:
writer = csv.writer(fp)
writer.writerow(['id', 'tested_num'])
for i, p in enumerate(preds):
writer.writerow([i, p])
#保存训练数据
def save_train_data(checkpoint_PATH):
print("Saving training data......")
torch.save({
'epoch': save_epoch + epoch + 1, 'state_dict': model.state_dict(),
'optimizer': optimizer.state_dict(), 'max_correct': max_correct},
checkpoint_PATH)
#加载训练数据
def load_train_data(checkpoint_PATH,model,optimizer):
print('loading train data......')
if os.path.isfile(checkpoint_PATH):
model_CKPT = torch.load(checkpoint_PATH)
model.load_state_dict(model_CKPT['state_dict'])
optimizer.load_state_dict(model_CKPT['optimizer'])
maxCorrect = model_CKPT['max_correct']
save_epoch = model_CKPT['epoch']
print("Successfully Loading file,last max_correct is ",maxCorrect,"%",'total epoch is',save_epoch)
return model, optimizer,maxCorrect,save_epoch
# 绘图函数
def plt(train_loss_list,test_loss_list):
epoch = np.arange(1, len(train_loss_list) + 1, 1)
train_loss_list = np.array(train_loss_list)
plt.plot(epoch, train_loss_list)
test_loss_list = np.array(test_loss_list)
plt.plot(epoch, test_loss_list)
plt.xlabel("Epoch")
plt.ylabel('Accuracy')
plt.grid()
plt.show()
if __name__ == '__main__':
max_correct = 0
epochs = 50
save_epoch = 0
start = time.time()
# 加载上次的训练数据
model,optimizer,max_correct,save_epoch = load_train_data("model_AD.tar",model,optimizer)
for epoch in range(epochs):
train(epoch,train_loader)
pred,correct = test(test_loader)
if correct > max_correct:
max_correct = correct
save_train_data("model_AD.tar")
print("The Highest correct rate: ", max_correct,"%")
#载入模型
model,optimizer,max_correct,save_epoch = load_train_data("model_AD.tar",model,optimizer)
#制作预测文件
pred, correct = test(test_loader)
save_pred(pred, 'pred.csv')
第四个模型ResNet经过13个epoch的训练之后的识别准确率可以高达99.43%，数据处理，优化算法的选择，模型的选择，各种超参数的调节缺一不可。