torch随手笔记

张量创建

#创建一个未初始化的5x3张量
x = torch.empty(5, 3)
#创建一个随机初始化的5x3张量
x = torch.rand(5, 3)
#创建一个5x3的0张量，类型为long
x = torch.zeros(5, 3, dtype=torch.long)
#直接从数组创建张量
x = torch.tensor([5.5, 3])
#从已有的张量创建相同维度的新张量，并且重新定义类型为float
x = torch.randn_like(x, dtype=torch.float)
#打印一个张量的维度
print(x.size())
#打印张量的第一列
print(x[:, 1])
#将一个4x4的张量resize成一个一维张量
x = torch.randn(4, 4)
y = x.view(16)
print(x.size(),y.size())
#将一个4x4的张量，resize成一个2x8的张量
y = x.view(2, 8)
z = x.view(-1, 8) # 确定一个维度，-1的维度会被自动计算
#从张量中取出数字
x = torch.randn(1)
print(x.item())
#从numpy数组创建张量
a = np.ones(5)
b = torch.from_numpy(a)
b = torch.tensor(a)
#张量运算
a=torch.tensor([[1,2],[3,4]])
b=torch.tensor([[1,2],[3,4]])
print(a*b)#点乘
print(torch.mul(a,b))#点乘
print(torch.mm(a,b))#矩阵乘

 自动微分

#新建一个张量，并设置requires_grad=True
x = torch.ones(2, 2, requires_grad=True)
y = x + 2
print(y.grad_fn) # y就多了一个AddBackward的对象
z = y * y * 3
out = z.mean()
#对out进行反向传播
out.backward()
#打印梯度d(out)/dx
print(x.grad)

 神经网络

import torch
import torch.nn as nn
import torch.nn.functional as F

class Net(nn.Module):
    def __init__(self):
        # super(Net, self).__init__()
        super().__init__()
        self.conv1 = nn.Conv2d(1, 6, 5)
        self.conv2 = nn.Conv2d(6, 16, 5)
        self.fc1 = nn.Linear(16 * 5 * 5, 120)  # 6*6 from image dimension
        self.fc2 = nn.Linear(120, 84)
        self.fc3 = nn.Linear(84, 10)

    def forward(self, x):
        # 31.完成input-S2，先卷积+relu，再2x2下采样
        x = F.max_pool2d(F.relu(self.conv1(x)), (2, 2))
        x = F.max_pool2d(F.relu(self.conv2(x)), 2) #卷积核方形时，可以只写一个维度
        # 33.将特征向量扁平成行向量
        x = x.view(-1, 16 * 5 * 5)
        # 34.使用fc1+relu
        x = F.relu(self.fc1(x))
        # 35.使用fc2+relu
        x = F.relu(self.fc2(x))
        # 36.使用fc3
        x = self.fc3(x)
        return x
net = Net()
# print(net)
# params = list(net.parameters())
# for i in range(len(params)):
#     print(params[i].size())

input=torch.randn(1,1,32,32)
target = torch.randn(10).view(1,-1)

optimizer = torch.optim.SGD(net.parameters(), lr=0.01)
optimizer.zero_grad()#初始化梯度

out=net(input)#前馈

criterion = nn.MSELoss()
loss=criterion(out,target)#计算损失
loss.backward()#计算梯度
optimizer.step()#更新权值

print('conv1.bias.grad before backward:',net.conv1.bias.grad)

模型训练保存 

训练 

for epoch in range(2): 

    running_loss = 0.0
    for i, data in enumerate(trainloader, 0):
        # 获取X,y对
        inputs, labels = data

        # 51.初始化梯度
        optimizer2.zero_grad()

        # 52.前馈
        outputs = net2(inputs)
        # 53.计算损失
        loss = criterion2(outputs, labels)
        # 54.计算梯度
        loss.backward()
        # 55.更新权值
        optimizer2.step()

        # 每2000个数据打印平均代价函数值
        running_loss += loss.item()
        if i % 2000 == 1999:    # print every 2000 mini-batches
            print('[%d, %5d] loss: %.3f' %
                  (epoch + 1, i + 1, running_loss / 2000))
            running_loss = 0.0

print('Finished Training')

保存

#保存训练好的模型
PATH = './cifar_net.pth'
torch.save(net.state_dict(), PATH)
#读取保存的模型
pretrained_net = torch.load(PATH)
#加载模型
net3 = Net()
net3.load_state_dict(pretrained_net)

参考代码：

https://www.kesci.com/home/workspace/project