pytorch 神经网络套路 实现多维输入特征的分类：MNIST手写数字分类（pytorch中的hello world）

1.数据集：

MNIST:手写数字 

下载方法：见代码段，dataset的下载属性download设置为True

2.模型：

本文中，去掉了64维特征的线性层，直接由128维特征降为10维，作为输出。

LOSS：采用CrossEntropyLoss，交叉熵

优化器：SGD随机梯度下降

3.python代码：

import torch
import torchvision
from torch import nn
from torch.nn import Linear, ReLU, CrossEntropyLoss
from torch.optim import SGD
from torch.utils.data import DataLoader
from torchvision import transforms
import matplotlib.pyplot as plt

# 定义计算设备，如果gpu可用就用cuda加速,否则使用cpu计算
device = ("cuda:0" if torch.cuda.is_available() else "cpu")

# 准备测试和验证集
train_dataset = torchvision.datasets.MNIST("./data_set/train_dataset", train=True, transform=transforms.Compose(
    [transforms.ToTensor(), transforms.Normalize(0.1307, 0.3081)]), download=True)

val_dataset = torchvision.datasets.MNIST("./data_set/val_dataset", train=True, transform=transforms.Compose(
    [transforms.ToTensor(), transforms.Normalize(0.1307, 0.3081)]), download=True)

train_loader = DataLoader(train_dataset, batch_size=64, shuffle=True)

val_loader = DataLoader(val_dataset, batch_size=64, shuffle=True)


# 建立模型
class model(nn.Module):
    def __init__(self):
        super(model, self).__init__()
        self.linear1 = Linear(784, 512, bias=True)
        self.linear2 = Linear(512, 256, bias=True)
        self.linear3 = Linear(256, 128, bias=True)
        self.linear4 = Linear(128, 10, bias=True)
        self.activate = ReLU()

    def forward(self, x):
        x = x.view(-1, 28 * 28)
        x = self.linear1(x)
        x = self.activate(x)
        x = self.linear2(x)
        x = self.activate(x)
        x = self.linear3(x)
        x = self.activate(x)
        x = self.linear4(x)
        return x


# 模型类实例化
my_model = model()
# gpu加速
my_model.to(device)

# 交叉熵损失函数
loss_cal = CrossEntropyLoss(size_average=True)
# gpu加速
loss_cal = loss_cal.to(device)

# 优化器
optimizer = SGD(my_model.parameters(), lr=0.01, momentum=0.1)


# 训练函数
def train():
    loss_sum = 0
    for i, data in enumerate(train_loader):
        imgs, labels = data
        # gpu加速
        imgs = imgs.to(device)
        labels = labels.to(device)
        # 前向计算
        outs = my_model(imgs)
        loss = loss_cal(outs, labels)
        loss_sum = loss_sum + loss.item()
        # 梯度清零
        optimizer.zero_grad()
        # 反向传播
        loss.backward()
        # 参数优化
        optimizer.step()
        if i % 50 == 49:
            loss_list.append(loss_sum / 50)
            print("loss:" + str(loss_sum / 50))
            loss_sum = 0


# 验证函数
def val(epoch):
    correct = 0
    total = 0
    with torch.no_grad():
        for i, data in enumerate(val_loader):
            imgs, labels = data
            imgs = imgs.to(device)
            labels = labels.to(device)
            outs = my_model(imgs)
            value, index = torch.max(outs.data, dim=1)
            total += labels.size(0)
            correct += (index == labels).sum().item()

    print("epoch:" + str(epoch) + "accuracy:" + str(correct / float(total)))


if __name__ == "__main__":

    loss_list = []
    for epoch in range(10):
        train()
        val(epoch)
    plt.figure()
    plt.plot(loss_list)
    plt.show()


# 部分输出结果
# loss:0.08146183107048273
# loss:0.09409760734066368
# loss:0.08619683284312486
# loss:0.0925342983007431
# loss:0.10231521874666213
# loss:0.09232219552621246
# loss:0.08993134327232838
# loss:0.08221664376556874
# loss:0.08144009610638023
# loss:0.08002457775175571
# loss:0.09039016446098685
# loss:0.09695547364652157
# loss:0.08471773650497198
# loss:0.09309148231521248
# loss:0.08477838601917029
# loss:0.08601660847663879
# loss:0.08170276714488864
# loss:0.0860577792301774
# epoch:8accuracy:0.9759166666666667
# loss:0.08367479223757983
# loss:0.08048359720036387
# loss:0.07391996223479509
# loss:0.08173186900094151
# loss:0.07706649962812662
# loss:0.08126413892954588
# loss:0.07230079263448715
# loss:0.08468491364270449
# loss:0.06976747298613191
# loss:0.07847631502896547
# loss:0.07585323289036751
# loss:0.08995127085596323
# loss:0.0836580672301352
# loss:0.07555182477459312
# loss:0.09279871977865696
# loss:0.06770527206361293
# loss:0.08244827814400196
# loss:0.07539539625868201
# epoch:9accuracy:0.9793333333333333

4.可视化结果：

 随着训练次数增加，loss逐渐减小并收敛。

5.以上均为个人学习pytorch基础入门中的基础，浅做记录，如有错误，请各位大佬批评指正！

6.关于问题描述和原理的部分图片参考刘老师的视频课件，本文也是课后作业的一部分，特此附上视频链接，《PyTorch深度学习实践》完结合集_哔哩哔哩_bilibili，希望大家都有所进步！