pytorch学习--使用m1 进行训练

import torch
#判断是否存在 gpu
torch.backends.mps.is_available()

True

if torch.backends.mps.is_available():
    mps_device = torch.device("mps")
    x = torch.ones(1, device=mps_device)
    print (x)
else:
    print ("MPS device not found.")

tensor([1.], device='mps:0')

加载数据

from torch.utils.data import Dataset
import os
from PIL import Image

class MyData(Dataset):
    def __init__(self,root_dir,label_dir):
        self.root_dir = root_dir
        self.label_dir = label_dir
        self.path = os.path.join(self.root_dir,self.label_dir)
        self.img_path = os.listdir(self.path)

    def __getitem__(self,idx):
        img_name = self.img_path[idx]
        img_item_path = os.path.join(self.root_dir, self.label_dir, img_name)
        img = Image.open(img_item_path)
        label = self.label_dir
        return img,label

    def __len__(self):
        return len(self.img_path)

ants_dataset = MyData("./hymenoptera_data/train/","ants")
ants_dataset.__len__()

124

TensorBoard 使用

from torch.utils.tensorboard import SummaryWriter
import numpy as np

img = Image.open("./hymenoptera_data/train/ants/0013035.jpg")
img_array = np.array(img)

# 创建 SummaryWriter 类的对象，指定日志文件夹
writer = SummaryWriter("logs")

writer.add_image("test", img_array, 1, dataformats="HWC")

for i in range(150):
    # step1: 使用 conda activate 环境名 激活环境
    # step2: 进入 日志文件 logs 的父文件夹路径下
    # step3：在终端输入 tensorboard --logdir=logs --port=端口号
    writer.add_scalar('y=x**2',i**2,i)

#writer.close()

transforms

from torchvision import transforms

#tensor
trans_totensor = transforms.ToTensor()
img_tensor = trans_totensor(img)
writer.add_image("img_tensor", img_tensor, 2)
type(img_tensor),img_tensor.shape

(torch.Tensor, torch.Size([3, 512, 768]))

#normalize 规范化
mean = [0.4,0.5,0.4]
std = [0.2,0.3,0.2]
trans_toNormalize = transforms.Normalize(mean,std)
img_normalize = trans_toNormalize(img_tensor)

writer.add_image("img_normalize", img_normalize, 3)
img_normalize[0,0,0]

tensor(-0.4314)

#Resize 改变大小 整体缩放扩大
trans_resize = transforms.Resize((512,512))
img_resize = trans_resize(img)
img_resize = trans_totensor(img_resize)

writer.add_image("img_resize", img_resize, 4)

img_resize.shape

torch.Size([3, 512, 512])

#Randomcrop 随机裁剪
trans_random = transforms.RandomCrop(512)
trans_compose = transforms.Compose([trans_random, trans_totensor])
img_random = trans_compose(img)

writer.add_image("img_random", img_random, 5)

img_random.shape

torch.Size([3, 512, 512])

torchvsion 中数据集的使用

import torchvision

dataset_transform = transforms.Compose([transforms.ToTensor()])

train_set = torchvision.datasets.CIFAR10("./dataset", train=True, transform=dataset_transform, download=True)
test_set = torchvision.datasets.CIFAR10("./dataset", train=False, transform=dataset_transform, download=True)

Files already downloaded and verified
Files already downloaded and verified

test_set[0]

img, labels = test_set[0]
display(img,labels)

tensor([[[0.6196, 0.6235, 0.6471,  ..., 0.5373, 0.4941, 0.4549],
         [0.5961, 0.5922, 0.6235,  ..., 0.5333, 0.4902, 0.4667],
         [0.5922, 0.5922, 0.6196,  ..., 0.5451, 0.5098, 0.4706],
         ...,
         [0.2667, 0.1647, 0.1216,  ..., 0.1490, 0.0510, 0.1569],
         [0.2392, 0.1922, 0.1373,  ..., 0.1020, 0.1137, 0.0784],
         [0.2118, 0.2196, 0.1765,  ..., 0.0941, 0.1333, 0.0824]],

        [[0.4392, 0.4353, 0.4549,  ..., 0.3725, 0.3569, 0.3333],
         [0.4392, 0.4314, 0.4471,  ..., 0.3725, 0.3569, 0.3451],
         [0.4314, 0.4275, 0.4353,  ..., 0.3843, 0.3725, 0.3490],
         ...,
         [0.4863, 0.3922, 0.3451,  ..., 0.3804, 0.2510, 0.3333],
         [0.4549, 0.4000, 0.3333,  ..., 0.3216, 0.3216, 0.2510],
         [0.4196, 0.4118, 0.3490,  ..., 0.3020, 0.3294, 0.2627]],

        [[0.1922, 0.1843, 0.2000,  ..., 0.1412, 0.1412, 0.1294],
         [0.2000, 0.1569, 0.1765,  ..., 0.1216, 0.1255, 0.1333],
         [0.1843, 0.1294, 0.1412,  ..., 0.1333, 0.1333, 0.1294],
         ...,
         [0.6941, 0.5804, 0.5373,  ..., 0.5725, 0.4235, 0.4980],
         [0.6588, 0.5804, 0.5176,  ..., 0.5098, 0.4941, 0.4196],
         [0.6275, 0.5843, 0.5176,  ..., 0.4863, 0.5059, 0.4314]]])



3

test_set.classes[labels]

'cat'

type(img)

torch.Tensor

dataloader 从 dataset 中读取数据，加载到 神经网络

from torch.utils.data import DataLoader

train_data = DataLoader(dataset=train_set,batch_size = 4,shuffle = True, num_workers = 0, drop_last = False)
test_data = DataLoader(dataset=test_set,batch_size = 4,shuffle = True, num_workers = 0, drop_last = False)

神经网络基本骨架 – nn.module 的使用

from torch import nn

class MyNet(nn.Module):
    def __init__(self):
        super().__init__()

    def forward(self, input):
        output = input + 1
        return output

#ceshi(data) & ceshi.forward(data) 等价，
#因为在class Module中调用了__call__函数，forward函数在call函数中调用了
ceshi = MyNet()
x = torch.tensor(2.0)
y = ceshi(x)
y

tensor(3.)

搭建实战

class MyNet2(nn.Module):
    def __init__(self):
        super().__init__()
        self.cov1 = nn.Conv2d(3,32,5,padding=2)
        self.maxpool1 = nn.MaxPool2d(2)
        self.cov2 = nn.Conv2d(32,32,5,padding=2)
        self.maxpool2 = nn.MaxPool2d(2)
        self.cov3 = nn.Conv2d(32,64,5,padding=2)
        self.maxpool3 = nn.MaxPool2d(2)
        self.flatten = nn.Flatten()
        self.liner1 = nn.Linear(1024,64)
        self.liner2 = nn.Linear(64,10)

    def forward(self,x):
        x = self.cov1(x)
        x = self.maxpool1(x)
        x = self.cov2(x)
        x = self.maxpool2(x)
        x = self.cov3(x)
        x = self.maxpool3(x)
        x = self.flatten(x)
        x = self.liner1(x)
        x = self.liner2(x)
        return x

type(train_data)

torch.utils.data.dataloader.DataLoader

for i, data in enumerate(train_data):
    print(data)
    break

ceshi_net = MyNet2()
ceshi_net

MyNet2(
  (cov1): Conv2d(3, 32, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2))
  (maxpool1): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
  (cov2): Conv2d(32, 32, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2))
  (maxpool2): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
  (cov3): Conv2d(32, 64, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2))
  (maxpool3): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
  (flatten): Flatten(start_dim=1, end_dim=-1)
  (liner1): Linear(in_features=1024, out_features=64, bias=True)
  (liner2): Linear(in_features=64, out_features=10, bias=True)
)

inputs = torch.ones((64,3,32,32),dtype=torch.float32)
inputs.shape

torch.Size([64, 3, 32, 32])

out_put = ceshi_net(inputs)
out_put.shape

torch.Size([64, 10])

损失函数与反向传播

inputs = torch.tensor([1,2,3] ,dtype = torch.float32)
targets = torch.tensor([1,2,5] ,dtype = torch.float32)

inputs = torch.reshape(inputs,(1,1,1,3))
targets = torch.reshape(targets,(1,1,1,3))
inputs,targets

(tensor([[[[1., 2., 3.]]]]), tensor([[[[1., 2., 5.]]]]))

loss = nn.L1Loss(reduction="sum")
result = loss(inputs, targets)
result

tensor(2.)

loss_mse = nn.MSELoss()
result_mse = loss_mse(inputs, targets)
result_mse

tensor(1.3333)

将数据集放入网络进行测试

lose_cro = nn.CrossEntropyLoss()
for data in train_data:
    imgs, labels = data
    outputs = ceshi_net(imgs)
    loss_values = lose_cro(outputs,labels)
    #进行反向传播时，使用的是损失值
    loss_values.backward()

优化器

#设置优化器 从 torch.optim 模块中导入
#第一个参数为 网络的各个层的权重、偏置等参数 使用方法  ceshi_net.parameters() --》ceshi_net 为你定义的网络
optimizer = torch.optim.SGD(ceshi_net.parameters(),lr=0.01)

for epoch in range(30):
    epoch_loss = 0.0
    for inputs, target in train_data:
        #将上次的梯度归0
        optimizer.zero_grad()
        output = ceshi_net(inputs)
        loss = lose_cro(output, target)
        #反向传播
        loss.backward()
        #更新参数
        optimizer.step()
        epoch_loss += loss

    print(f'第 {epoch} 次的损失值为 {epoch_loss}')

#输出每一层的 权重参数
for param in ceshi_net.parameters():
    print(param)

#实验自己的网络模型能否添加新的网络层
ceshi_net.add_module("update_linear",nn.Linear(10,20))
ceshi_net

MyNet2(
  (cov1): Conv2d(3, 32, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2))
  (maxpool1): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
  (cov2): Conv2d(32, 32, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2))
  (maxpool2): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
  (cov3): Conv2d(32, 64, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2))
  (maxpool3): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
  (flatten): Flatten(start_dim=1, end_dim=-1)
  (liner1): Linear(in_features=1024, out_features=64, bias=True)
  (liner2): Linear(in_features=64, out_features=10, bias=True)
  (update_linear): Linear(in_features=10, out_features=20, bias=True)
)

现有网络模型的使用及修改

# 使用 torchvision.models 中 现有的网络模型
vgg16_false = torchvision.models.vgg16(progress=False)
vgg16_true = torchvision.models.vgg16(progress=True)

vgg16_false

VGG(
  (features): Sequential(
    (0): Conv2d(3, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (1): ReLU(inplace=True)
    (2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (3): ReLU(inplace=True)
    (4): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
    (5): Conv2d(64, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (6): ReLU(inplace=True)
    (7): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (8): ReLU(inplace=True)
    (9): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
    (10): Conv2d(128, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (11): ReLU(inplace=True)
    (12): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (13): ReLU(inplace=True)
    (14): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (15): ReLU(inplace=True)
    (16): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
    (17): Conv2d(256, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (18): ReLU(inplace=True)
    (19): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (20): ReLU(inplace=True)
    (21): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (22): ReLU(inplace=True)
    (23): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
    (24): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (25): ReLU(inplace=True)
    (26): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (27): ReLU(inplace=True)
    (28): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (29): ReLU(inplace=True)
    (30): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
  )
  (avgpool): AdaptiveAvgPool2d(output_size=(7, 7))
  (classifier): Sequential(
    (0): Linear(in_features=25088, out_features=4096, bias=True)
    (1): ReLU(inplace=True)
    (2): Dropout(p=0.5, inplace=False)
    (3): Linear(in_features=4096, out_features=4096, bias=True)
    (4): ReLU(inplace=True)
    (5): Dropout(p=0.5, inplace=False)
    (6): Linear(in_features=4096, out_features=1000, bias=True)
  )
)

# 加载数据集
dataset_transform = transforms.Compose([transforms.ToTensor()])

train_set = torchvision.datasets.CIFAR10("./dataset", train=True, transform=dataset_transform, download=True)
test_set = torchvision.datasets.CIFAR10("./dataset", train=False, transform=dataset_transform, download=True)

Files already downloaded and verified
Files already downloaded and verified

# 数据集是 10 分类，vgg16 最后一层是1000，表示是1000分类  无法直接使用，需要调整
len(train_set.classes)

10

# 方式一 在最后一层后面 添加新的一层网络
vgg16_true.add_module("add_linear",nn.Linear(1000,10))
vgg16_true

VGG(
  (features): Sequential(
    (0): Conv2d(3, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (1): ReLU(inplace=True)
    (2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (3): ReLU(inplace=True)
    (4): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
    (5): Conv2d(64, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (6): ReLU(inplace=True)
    (7): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (8): ReLU(inplace=True)
    (9): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
    (10): Conv2d(128, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (11): ReLU(inplace=True)
    (12): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (13): ReLU(inplace=True)
    (14): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (15): ReLU(inplace=True)
    (16): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
    (17): Conv2d(256, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (18): ReLU(inplace=True)
    (19): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (20): ReLU(inplace=True)
    (21): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (22): ReLU(inplace=True)
    (23): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
    (24): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (25): ReLU(inplace=True)
    (26): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (27): ReLU(inplace=True)
    (28): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (29): ReLU(inplace=True)
    (30): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
  )
  (avgpool): AdaptiveAvgPool2d(output_size=(7, 7))
  (classifier): Sequential(
    (0): Linear(in_features=25088, out_features=4096, bias=True)
    (1): ReLU(inplace=True)
    (2): Dropout(p=0.5, inplace=False)
    (3): Linear(in_features=4096, out_features=4096, bias=True)
    (4): ReLU(inplace=True)
    (5): Dropout(p=0.5, inplace=False)
    (6): Linear(in_features=4096, out_features=1000, bias=True)
  )
  (add_linear): Linear(in_features=1000, out_features=10, bias=True)
)

# 方式二 修改最后一层后网络
vgg16_false.classifier[6] = nn.Linear(4096,10)
vgg16_false

VGG(
  (features): Sequential(
    (0): Conv2d(3, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (1): ReLU(inplace=True)
    (2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (3): ReLU(inplace=True)
    (4): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
    (5): Conv2d(64, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (6): ReLU(inplace=True)
    (7): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (8): ReLU(inplace=True)
    (9): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
    (10): Conv2d(128, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (11): ReLU(inplace=True)
    (12): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (13): ReLU(inplace=True)
    (14): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (15): ReLU(inplace=True)
    (16): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
    (17): Conv2d(256, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (18): ReLU(inplace=True)
    (19): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (20): ReLU(inplace=True)
    (21): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (22): ReLU(inplace=True)
    (23): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
    (24): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (25): ReLU(inplace=True)
    (26): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (27): ReLU(inplace=True)
    (28): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (29): ReLU(inplace=True)
    (30): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
  )
  (avgpool): AdaptiveAvgPool2d(output_size=(7, 7))
  (classifier): Sequential(
    (0): Linear(in_features=25088, out_features=4096, bias=True)
    (1): ReLU(inplace=True)
    (2): Dropout(p=0.5, inplace=False)
    (3): Linear(in_features=4096, out_features=4096, bias=True)
    (4): ReLU(inplace=True)
    (5): Dropout(p=0.5, inplace=False)
    (6): Linear(in_features=4096, out_features=10, bias=True)
  )
)

网络模型的 保存与读取

# 模型保存 方式一  ==>保存 模型结构和模型参数
torch.save(vgg16_false, "./model_save/vgg16_false_method1.pth")

# 模型加载 方式一
vgg16 = torch.load("./model_save/vgg16_false_method1.pth")
vgg16

VGG(
  (features): Sequential(
    (0): Conv2d(3, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (1): ReLU(inplace=True)
    (2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (3): ReLU(inplace=True)
    (4): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
    (5): Conv2d(64, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (6): ReLU(inplace=True)
    (7): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (8): ReLU(inplace=True)
    (9): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
    (10): Conv2d(128, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (11): ReLU(inplace=True)
    (12): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (13): ReLU(inplace=True)
    (14): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (15): ReLU(inplace=True)
    (16): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
    (17): Conv2d(256, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (18): ReLU(inplace=True)
    (19): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (20): ReLU(inplace=True)
    (21): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (22): ReLU(inplace=True)
    (23): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
    (24): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (25): ReLU(inplace=True)
    (26): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (27): ReLU(inplace=True)
    (28): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (29): ReLU(inplace=True)
    (30): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
  )
  (avgpool): AdaptiveAvgPool2d(output_size=(7, 7))
  (classifier): Sequential(
    (0): Linear(in_features=25088, out_features=4096, bias=True)
    (1): ReLU(inplace=True)
    (2): Dropout(p=0.5, inplace=False)
    (3): Linear(in_features=4096, out_features=4096, bias=True)
    (4): ReLU(inplace=True)
    (5): Dropout(p=0.5, inplace=False)
    (6): Linear(in_features=4096, out_features=10, bias=True)
  )
)

# 模型保存 方式二  ==>保存 模型参数
torch.save(vgg16_false.state_dict(), "./model_save/vgg16_false_method2.pth")

# 模型加载 方式二 ==>直接加载，显示的是 模型参数
vgg16_2 = torch.load("./model_save/vgg16_false_method2.pth")
vgg16_2

OrderedDict([('features.0.weight',
              tensor([[[[-0.0168, -0.0325, -0.0205],
                        [-0.0935, -0.0323,  0.0423],
                        [-0.0430,  0.0772, -0.0303]],
              
                       [[-0.0483,  0.0410, -0.0169],
                        [-0.0571, -0.0028,  0.0750],
                        [ 0.0327, -0.0050,  0.0498]],
              
                       [[ 0.0227,  0.0167,  0.0986],
                        [-0.0186,  0.0407, -0.0542],
                        [-0.0003, -0.1219, -0.0128]]],
              
              
                      [[[ 0.0837,  0.0252,  0.0755],
                        [-0.0458, -0.0566,  0.0238],
                        [-0.0009, -0.0628,  0.0475]],
              
                       [[-0.0773, -0.0765, -0.0098],
                        [ 0.0113,  0.0285,  0.0100],
                        [ 0.0049,  0.0168, -0.0904]],
              
                       [[ 0.0113, -0.0757,  0.1109],
                        [ 0.0532, -0.0509,  0.0531],
                        [-0.0180,  0.0739,  0.0259]]],
              
              
                      [[[-0.0281,  0.0229,  0.0258],
                        [ 0.0484,  0.1161, -0.0560],
                        [ 0.0428, -0.0409,  0.1285]],
              
                       [[-0.0033, -0.0482,  0.1041],
                        [ 0.0417,  0.1011, -0.0453],
                        [-0.0313, -0.0538,  0.0357]],
              
                       [[-0.0417,  0.0482,  0.0200],
                        [-0.0216,  0.1091, -0.0519],
                        [ 0.0002,  0.0265, -0.0066]]],
              
              
                      ...,
       
                      [-0.0115,  0.0140,  0.0149,  ...,  0.0089,  0.0155, -0.0049],
                      [ 0.0082,  0.0035,  0.0130,  ...,  0.0147, -0.0052, -0.0023],
                      [ 0.0055,  0.0116,  0.0001,  ...,  0.0019, -0.0021, -0.0098]])),
             ('classifier.6.bias',
              tensor([-0.0146,  0.0148, -0.0090,  0.0127, -0.0008,  0.0108, -0.0094,  0.0004,
                       0.0076, -0.0084]))])

#需要建立对应的 网络模型结构，使用 网络模型结构 加载 网络模型参数
vgg16_false_2 = torchvision.models.vgg16(progress=False)
vgg16_false_2.classifier[6] = nn.Linear(4096,10)
vgg16_false_2.load_state_dict(vgg16_2)
vgg16_false_2

VGG(
  (features): Sequential(
    (0): Conv2d(3, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (1): ReLU(inplace=True)
    (2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (3): ReLU(inplace=True)
    (4): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
    (5): Conv2d(64, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (6): ReLU(inplace=True)
    (7): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (8): ReLU(inplace=True)
    (9): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
    (10): Conv2d(128, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (11): ReLU(inplace=True)
    (12): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (13): ReLU(inplace=True)
    (14): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (15): ReLU(inplace=True)
    (16): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
    (17): Conv2d(256, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (18): ReLU(inplace=True)
    (19): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (20): ReLU(inplace=True)
    (21): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (22): ReLU(inplace=True)
    (23): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
    (24): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (25): ReLU(inplace=True)
    (26): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (27): ReLU(inplace=True)
    (28): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (29): ReLU(inplace=True)
    (30): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
  )
  (avgpool): AdaptiveAvgPool2d(output_size=(7, 7))
  (classifier): Sequential(
    (0): Linear(in_features=25088, out_features=4096, bias=True)
    (1): ReLU(inplace=True)
    (2): Dropout(p=0.5, inplace=False)
    (3): Linear(in_features=4096, out_features=4096, bias=True)
    (4): ReLU(inplace=True)
    (5): Dropout(p=0.5, inplace=False)
    (6): Linear(in_features=4096, out_features=10, bias=True)
  )
)

模型训练套路 -完整的训练过程

from torch.utils.data import DataLoader
# 加载数据集
dataset_transform = transforms.Compose([transforms.ToTensor()])

train_set = torchvision.datasets.CIFAR10("./dataset", train=True, transform=dataset_transform, download=True)
test_set = torchvision.datasets.CIFAR10("./dataset", train=False, transform=dataset_transform, download=True)

train_data = DataLoader(dataset=train_set,batch_size = 64,shuffle = True, num_workers = 0, drop_last = False)
test_data = DataLoader(dataset=test_set,batch_size = 64,shuffle = True, num_workers = 0, drop_last = False)

#学习率 1e-2 = 0.01
learn_rate = 1e-2

Files already downloaded and verified
Files already downloaded and verified

from torchvision.models import vgg16
#定义网络
#ceshi_vgg16 = vgg16(progress=False)
#ceshi_vgg16.add_module("add_linear",nn.Linear(1000,10))
ceshi_vgg16 = MyNet2()
print(ceshi_vgg16)

MyNet2(
  (cov1): Conv2d(3, 32, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2))
  (maxpool1): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
  (cov2): Conv2d(32, 32, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2))
  (maxpool2): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
  (cov3): Conv2d(32, 64, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2))
  (maxpool3): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
  (flatten): Flatten(start_dim=1, end_dim=-1)
  (liner1): Linear(in_features=1024, out_features=64, bias=True)
  (liner2): Linear(in_features=64, out_features=10, bias=True)
)

import time
start = time.time()
#定义优化器
optimizer = torch.optim.SGD(ceshi_vgg16.parameters(),lr=learn_rate)
#定义损失函数
loss = nn.CrossEntropyLoss()
#定义迭代次数
epochs = 5

#开始循环训练
for epoch in range(epochs):
    print(f"第 {epoch + 1} 次迭代训练开始....")
    epoch_train_loss = 0
    epoch_test_loss = 0

    #如果 网络结构中存在 dropout 层，需要添加  ceshi_vgg16.train() ==> ceshi_vgg16 为你的网络模型
    ceshi_vgg16.train()
    for data in train_data:
        inputs,targets = data
        outputs = ceshi_vgg16(inputs)
        
        train_loss = loss(outputs,targets)

        #梯度归 0
        optimizer.zero_grad()
        #反向传播
        train_loss.backward()
        #梯度更新
        optimizer.step()

        epoch_train_loss += train_loss

    #如果 网络结构中存在 dropout 层，需要添加  ceshi_vgg16.eval() ==> ceshi_vgg16 为你的网络模型
    ceshi_vgg16.eval()
    with torch.no_grad():
        for data in test_data:
            test_inputs,test_targets = data
            test_outputs = ceshi_vgg16(test_inputs)
            test_loss = loss(test_outputs,test_targets)

            epoch_test_loss += test_loss

    print(f"第 {epoch + 1} 次迭代训练,训练集损失为： {epoch_train_loss} ，测试集损失为： {epoch_test_loss}")

end = time.time()
print(f"cpu 训练时间为：{end - start}")

第 1 次迭代训练开始....
第 1 次迭代训练,训练集损失为： 1688.6795654296875 ，测试集损失为： 303.96173095703125
第 2 次迭代训练开始....
第 2 次迭代训练,训练集损失为： 1426.980712890625 ，测试集损失为： 268.0997314453125
第 3 次迭代训练开始....
第 3 次迭代训练,训练集损失为： 1285.7080078125 ，测试集损失为： 254.33773803710938
第 4 次迭代训练开始....
第 4 次迭代训练,训练集损失为： 1196.3980712890625 ，测试集损失为： 241.3218231201172
第 5 次迭代训练开始....
第 5 次迭代训练,训练集损失为： 1128.6549072265625 ，测试集损失为： 231.64186096191406
cpu 训练时间为：326.99496698379517

GPU 训练

from torchvision.models import vgg16
#定义网络
#ceshi_vgg16 = vgg16(progress=False)
#ceshi_vgg16.add_module("add_linear",nn.Linear(1000,10))
ceshi_vgg16 = MyNet2()
print(ceshi_vgg16)

MyNet2(
  (cov1): Conv2d(3, 32, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2))
  (maxpool1): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
  (cov2): Conv2d(32, 32, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2))
  (maxpool2): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
  (cov3): Conv2d(32, 64, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2))
  (maxpool3): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
  (flatten): Flatten(start_dim=1, end_dim=-1)
  (liner1): Linear(in_features=1024, out_features=64, bias=True)
  (liner2): Linear(in_features=64, out_features=10, bias=True)
)

# 获取 mac m1芯片（即GPU） 
device=torch.device("mps")

import time
start = time.time()

#定义优化器
optimizer = torch.optim.SGD(ceshi_vgg16.parameters(),lr=learn_rate)
#定义损失函数
loss = nn.CrossEntropyLoss()
#将损失函数放到 GPU 上
loss = loss.to(device)
#定义迭代次数
epochs = 20

#将 网络模型 放到 GPU 上
ceshi_vgg16 = ceshi_vgg16.to(device)

#开始循环训练
for epoch in range(epochs):
    print(f"第 {epoch + 1} 次迭代训练开始....")
    epoch_train_loss = 0
    epoch_test_loss = 0

    #如果 网络结构中存在 dropout 层，需要添加  ceshi_vgg16.train() ==> ceshi_vgg16 为你的网络模型
    ceshi_vgg16.train()
    for data in train_data:
        inputs,targets = data

        #将 数据 放到 GPU 上
        inputs = inputs.to(device)
        targets = targets.to(device)
        
        outputs = ceshi_vgg16(inputs)
        
        train_loss = loss(outputs,targets)

        #梯度归 0
        optimizer.zero_grad()
        #反向传播
        train_loss.backward()
        #梯度更新
        optimizer.step()

        epoch_train_loss += train_loss

    #如果 网络结构中存在 dropout 层，需要添加  ceshi_vgg16.eval() ==> ceshi_vgg16 为你的网络模型
    ceshi_vgg16.eval()
    with torch.no_grad():
        for data in test_data:
            test_inputs,test_targets = data
            #将 数据 放到 GPU 上
            test_inputs = test_inputs.to(device)
            test_targets = test_targets.to(device)
            
            test_outputs = ceshi_vgg16(test_inputs)
            test_loss = loss(test_outputs,test_targets)

            epoch_test_loss += test_loss

    print(f"第 {epoch + 1} 次迭代训练,训练集损失为： {epoch_train_loss} ，测试集损失为： {epoch_test_loss}")

end = time.time()
print(f"gpu 训练时间为：{end - start}")

第 1 次迭代训练开始....
第 1 次迭代训练,训练集损失为： 820.5022583007812 ，测试集损失为： 189.2407684326172
第 2 次迭代训练开始....
第 2 次迭代训练,训练集损失为： 790.6588134765625 ，测试集损失为： 175.98353576660156
第 3 次迭代训练开始....
第 3 次迭代训练,训练集损失为： 762.98876953125 ，测试集损失为： 179.2063751220703
第 4 次迭代训练开始....
第 4 次迭代训练,训练集损失为： 739.4859008789062 ，测试集损失为： 169.62254333496094
第 5 次迭代训练开始....
第 5 次迭代训练,训练集损失为： 716.714599609375 ，测试集损失为： 198.77293395996094
第 6 次迭代训练开始....
第 6 次迭代训练,训练集损失为： 695.8759155273438 ，测试集损失为： 165.21372985839844
第 7 次迭代训练开始....
第 7 次迭代训练,训练集损失为： 678.6303100585938 ，测试集损失为： 169.66461181640625
第 8 次迭代训练开始....
第 8 次迭代训练,训练集损失为： 657.269775390625 ，测试集损失为： 179.4607391357422
第 9 次迭代训练开始....
第 9 次迭代训练,训练集损失为： 639.9208984375 ，测试集损失为： 164.27601623535156
第 10 次迭代训练开始....
第 10 次迭代训练,训练集损失为： 623.4369506835938 ，测试集损失为： 166.43560791015625
第 11 次迭代训练开始....
第 11 次迭代训练,训练集损失为： 606.757568359375 ，测试集损失为： 189.4309539794922
第 12 次迭代训练开始....
第 12 次迭代训练,训练集损失为： 587.788330078125 ，测试集损失为： 181.8220672607422
第 13 次迭代训练开始....
第 13 次迭代训练,训练集损失为： 572.0140380859375 ，测试集损失为： 151.99435424804688
第 14 次迭代训练开始....
第 14 次迭代训练,训练集损失为： 556.7083129882812 ，测试集损失为： 174.7952880859375
第 15 次迭代训练开始....
第 15 次迭代训练,训练集损失为： 544.226318359375 ，测试集损失为： 173.84246826171875
第 16 次迭代训练开始....
第 16 次迭代训练,训练集损失为： 527.2423095703125 ，测试集损失为： 165.7640380859375
第 17 次迭代训练开始....
第 17 次迭代训练,训练集损失为： 512.9745483398438 ，测试集损失为： 173.00523376464844
第 18 次迭代训练开始....
第 18 次迭代训练,训练集损失为： 501.0533142089844 ，测试集损失为： 196.33889770507812
第 19 次迭代训练开始....
第 19 次迭代训练,训练集损失为： 487.0603332519531 ，测试集损失为： 167.21218872070312
第 20 次迭代训练开始....
第 20 次迭代训练,训练集损失为： 473.097412109375 ，测试集损失为： 173.1332550048828
gpu 训练时间为：124.73064494132996