PyTorch学习笔记（一）

import torch 
import torchvision
import torch.nn as nn
import torchvision.transforms as transforms
import numpy as np

1. Basic autograd example 1

# 创建tensors
x = torch.tensor(1.,requires_grad=True)
w = torch.tensor(2.,requires_grad=True)
b = torch.tensor(3.,requires_grad=True)

#  构建计算图
y = w * x + b # y=2x+3

# 计算梯度
y.backward()

# 打印梯度
print(x.grad)
print(w.grad)
print(b.grad)

tensor(2.)
tensor(1.)
tensor(1.)

2. Basic autograd example 2

# 创建tensors (10,3) (10,2)
x = torch.randn(10,3)
y = torch.randn(10,2)
print(x)
print(y)

tensor([[ 0.0191, -0.0627,  0.9497],
        [ 0.4595, -0.2088, -0.2721],
        [-0.2952,  0.0117, -0.2056],
        [ 0.1166,  0.3428,  0.0982],
        [-0.0378,  0.5100,  0.1092],
        [ 0.7950, -0.9665, -0.4156],
        [ 0.0040,  0.4486, -0.0933],
        [ 0.4311, -1.4221, -0.9666],
        [-0.4988, -0.6366,  1.2200],
        [ 0.6022, -0.5120, -0.0679]])
tensor([[-0.3624,  0.3539],
        [-0.2107,  1.1979],
        [-0.0265, -0.1485],
        [-2.2705, -0.9377],
        [-0.2282, -0.4204],
        [-0.4186,  0.6560],
        [-0.6858, -1.7737],
        [-0.4479,  0.8466],
        [ 0.3069, -0.4114],
        [-0.0926, -1.4976]])

# 构建一个全连接层
linear = nn.Linear(3,2)
print(linear)
print('w:',linear.weight)
print('b:',linear.bias)

Linear(in_features=3, out_features=2, bias=True)
w: Parameter containing:
tensor([[ 0.2266,  0.4908, -0.2475],
        [ 0.2061,  0.5138,  0.1486]], requires_grad=True)
b: Parameter containing:
tensor([-0.3633, -0.5637], requires_grad=True)

# 构建损失函数和优化器
criterion = nn.MSELoss()
optimizer = torch.optim.SGD(linear.parameters(),lr=0.01)

# Forward pass
pred = linear(x)

# 计算损失
loss = criterion(pred,y)
print('loss:',loss.item())

loss: 1.0909489393234253

# Backward pass.
loss.backward()

# 打印梯度
print('dL/dw:',linear.weight.grad)
print('dL/db:',linear.bias.grad)

dL/dw: tensor([[ 0.0548,  0.2640, -0.1351],
        [-0.2189,  0.5867,  0.2022]])
dL/db: tensor([-0.0148, -0.4402])

# 进行一次梯度下降
optimizer.step()

# 打印一次梯度下降后的损失
pred = linear(x)
loss = criterion(pred,y)
print('loss after 1 step optimization:',loss.item())

loss after 1 step optimization: 1.0838005542755127

3. Loading data from numpy

# 创建数组
x = np.array(([1,2],[3,4]))

# 将数组转化为tensor
y = torch.from_numpy(x)

# 将tensor转化为数组
z = y.numpy()

4. Input pipeline

# 下载和构造CIFAR10数据集
train_dataset = torchvision.datasets.CIFAR10(root='../../data/',
                                            train=True,
                                            transform=transforms.ToTensor(),
                                            download=True)

Downloading https://www.cs.toronto.edu/~kriz/cifar-10-python.tar.gz to ../../data/cifar-10-python.tar.gz


26.4%IOPub message rate exceeded.
The notebook server will temporarily stop sending output
to the client in order to avoid crashing it.
To change this limit, set the config variable
`--NotebookApp.iopub_msg_rate_limit`.

Current values:
NotebookApp.iopub_msg_rate_limit=1000.0 (msgs/sec)
NotebookApp.rate_limit_window=3.0 (secs)

90.7%IOPub message rate exceeded.
The notebook server will temporarily stop sending output
to the client in order to avoid crashing it.
To change this limit, set the config variable
`--NotebookApp.iopub_msg_rate_limit`.

Current values:
NotebookApp.iopub_msg_rate_limit=1000.0 (msgs/sec)
NotebookApp.rate_limit_window=3.0 (secs)

100.0%


Extracting ../../data/cifar-10-python.tar.gz to ../../data/

# 读取数据 one data pair
image,label = train_dataset[0]
print(image.size())
print(label)

torch.Size([3, 32, 32])
6

# Data loader
train_loader = torch.utils.data.DataLoader(dataset=train_dataset,
                                          batch_size=64,
                                          shuffle=True)

# 迭代开始后，队列和线程将从文件里读取数据
data_iter = iter(train_loader)

# Mini-batch images and labels
images,labels = data_iter.next()

# 实际使用data loader如下
for images,labels in train_loader:
    # 训练代码
    pass

5. Input pipeline for custom dataset

#  构建自定义数据集
class CustomDataset(torch.utils.data.Dataset):
    def __init__(self):
        # 初始化文件路径或者文件名
        pass
    def __getitem__(self,index):
        # 1.从文件中读取数据
        # 2.处理数据
        # 3.返回数据对（图像和标签）
        pass
    def __len__(self):
        # 改变数据集总的大小
        return 0

# 使用构建好的data loader
custom_dataset = CustomDataset()
train_loader = torch.utils.data.DataLoader(dataset=custom_dataset,
                                           batch_size=64, 
                                           shuffle=True)

6. Pretrained model

# 下载和导入预训练模型 ResNet-18
resnet = torchvision.models.resnet18(pretrained=True)

Downloading: "https://download.pytorch.org/models/resnet18-f37072fd.pth" to C:\Users\wydxr/.cache\torch\hub\checkpoints\resnet18-f37072fd.pth
100.0%

# If you want to finetune only the top layer of the model, set as below.
for param in resnet.parameters():
    param.requires_grad = False

# 替换掉最上层做微调
resnet.fc = nn.Linear(resnet.fc.in_features,100) 

# 前向传播
images = torch.randn(64,3,224,224)
outputs = resnet(images)
print(outputs.size())

torch.Size([64, 100])

7. Save and load the model

# 保存和导入整个模型
torch.save(resnet,'model.ckpt')
model = torch.load('model.ckpt')

# 保存和导入模型的参数（推荐）
torch.save(resnet.state_dict(),'params.ckpt')
resnet.load_state_dict(torch.load('params.ckpt'))

<All keys matched successfully>