116 lines
4.0 KiB
Python
116 lines
4.0 KiB
Python
import torch
|
||
import torchvision.datasets
|
||
from matplotlib import pyplot as plt
|
||
from torch import nn
|
||
from torch.utils.data import DataLoader
|
||
|
||
batch_size=10
|
||
train_data=torchvision.datasets.MNIST(root="./data",train=True,download=True,transform=torchvision.transforms.ToTensor())
|
||
test_data=torchvision.datasets.MNIST(root="./data",train=False,download=True,transform=torchvision.transforms.ToTensor())
|
||
|
||
|
||
train_dataloader=DataLoader(dataset=train_data,batch_size=batch_size)
|
||
test_dataloader=DataLoader(dataset=test_data,batch_size=batch_size)
|
||
|
||
|
||
# #输出数据集中的第一个图片
|
||
# plt.imshow(train_data.data[0].numpy(),cmap='gray')
|
||
# plt.title('%i' % train_data.targets[0])
|
||
# plt.show()
|
||
|
||
|
||
class AlexNet(nn.Module):
|
||
def __init__(self):
|
||
super().__init__()
|
||
self.conv1=nn.Sequential(
|
||
nn.Conv2d(in_channels=1,out_channels=96,kernel_size=3,stride=1,padding=1),#28*28
|
||
nn.ReLU(),
|
||
#2.改 3通道
|
||
#3. 写net 传函数 和 新模型写类区别
|
||
#4.用batch
|
||
nn.MaxPool2d(kernel_size=3,stride=1)#26*26
|
||
)
|
||
self.conv2=nn.Sequential(
|
||
nn.Conv2d(in_channels=96,out_channels=256,kernel_size=5,padding=2),#26*26
|
||
nn.ReLU(),
|
||
nn.MaxPool2d(kernel_size=3,stride=1)#24*24
|
||
)
|
||
self.conv3=nn.Sequential(
|
||
nn.Conv2d(in_channels=256,out_channels=384,kernel_size=3,padding=1),
|
||
nn.ReLU(),
|
||
nn.Conv2d(in_channels=384,out_channels=384,kernel_size=3,padding=1),
|
||
nn.ReLU(),
|
||
nn.Conv2d(in_channels=384,out_channels=256,kernel_size=3,padding=1),
|
||
nn.ReLU(),
|
||
nn.MaxPool2d(kernel_size=3,stride=1)#22*22
|
||
)
|
||
self.flatten=nn.Flatten()
|
||
self.end=nn.Sequential(
|
||
nn.Linear(in_features=256*22*22,out_features=256),
|
||
nn.ReLU(),
|
||
nn.Linear(in_features=256,out_features=10)
|
||
)
|
||
def forward(self,x):
|
||
x=self.conv1(x)
|
||
x=self.conv2(x)
|
||
x=self.conv3(x)
|
||
x=self.flatten(x)
|
||
x=self.end(x)
|
||
return x
|
||
|
||
|
||
net=AlexNet()
|
||
loss_cross=nn.CrossEntropyLoss()
|
||
|
||
optim=torch.optim.Adam(net.parameters(),lr=0.001)
|
||
|
||
def train(net,train_data):
|
||
for data,targets in train_data:
|
||
#print("targets:",targets)
|
||
#print("data.shape:",data.shape)
|
||
#print("targets.shape:",targets.shape)
|
||
#print("targets.dty24ALexNet.pype",targets.dtype)
|
||
#data=data.reshape(10,1,28,28) #1,28,28->1,1,28,28 ?
|
||
data=net(data)
|
||
|
||
#计算每次训练的准确率
|
||
acc=(data.argmax(1)==targets).sum() #用sum() 和 sum(0)都可以 || 加()的是函数 不加()的是属性
|
||
#print(acc.dtype)
|
||
accury=acc/batch_size
|
||
#print("正确数：",acc,"正确率",accury)
|
||
print("正确数:{},正确率:{}".format(acc,accury))#跑的时候，除10 有 0.0000几的误差????
|
||
loss= loss_cross(data,targets)
|
||
#print("loss:",loss) #loss值代表？??
|
||
print("loss:{}".format(loss))
|
||
loss.backward()
|
||
optim.step()
|
||
optim.zero_grad()
|
||
return net
|
||
|
||
def yanzheng(net,test_data):
|
||
net.eval()
|
||
acc=0
|
||
sum=batch_size
|
||
with torch.no_grad():
|
||
for data,targets in test_data:
|
||
data=net(data)
|
||
acc=(data.argmax(1)==targets).sum()
|
||
accury = acc / batch_size
|
||
print("正确数：",acc," ","正确率",accury)
|
||
|
||
#训练数据
|
||
#net=train(net,train_dataloader) #不返回也没事 训练之后的参数会保留,(只要是在这一次程序中运行的)
|
||
|
||
#保存模型 此方法需要保证可以找到自己定义的class AlexNet模型 （直接加载模型应该也可以吧 eg:VGG） ||我认为小土堆的第二种方法 也需要保证运行的程序中有模型，无论是加载别人定义好的 还是 自己引入（from P26_model_save import *）/加上代码 class 模型
|
||
#torch.save(net,"ALexNet_02.path")
|
||
|
||
#加载模型
|
||
net_loading=torch.load("ALexNet_02.path")
|
||
|
||
#测试数据
|
||
yanzheng(net_loading,test_dataloader)
|
||
|
||
|
||
|
||
|