PyTorch学习系列教程：构建一个深度学习模型需要哪几步？

class MyDataset(Dataset):
    def __init__(self, x, y):
        super().__init__()
        ……

    def __getitem__(self):
        return ……

    def __len__(self):
        return ……

my_dataset = TensorDataset(tenso_x, tensor_y)

dataloader = DataLoader(MyDataset, batch_size=128, shuffle=True)

from sklearn.datasets import load_digits
from sklearn.model_selection import train_test_split
from sklearn.ensemble import RandomForestClassifier
X, y = load_digits(return_X_y=True)
X_train, X_test, y_train, y_test = train_test_split(X, y)
rf = RandomForestClassifier().fit(X_train, y_train)
rf.score(X_test, y_test)
### 输出：0.9688888888888889

import torch
from torch.utils.data import TensorDataset, DataLoader

X_train_tensor = torch.Tensor(X_train)
y_train_tensor = torch.Tensor(y_train).long()  # 主要标签需要用整数形式，否则后续用于计算交叉熵损失时报错
dataset = TensorDataset(X_train_tensor, y_train_tensor)  # 直接调用TensorDataset加以包裹使用
dataloader = DataLoader(dataset, batch_size=128, shuffle=True)  # 每128个样本为一个batch，训练时设为随机

X_test_tensor = torch.Tensor(X_test)  # 测试集只需转化为tensor即可
y_test_tensor = torch.Tensor(y_test).long()

from torch import nn, optim
class Model(nn.Module):  # 继承Module基类
    def __init__(self, n_input=64, n_hidden=32, n_ouput=10):
        # 定义一个含有单隐藏层的全连接网络，其中输入64为手写数字数据集的特征数，输出10为类别数，隐藏层神经元数量设置32
        super().__init__()
        # 使用全连接层和ReLU激活函数搭建网络模型
        self.dnn = nn.Sequential(
            nn.Linear(n_input, n_hidden),
            nn.ReLU(),
            nn.Linear(n_hidden, n_output)
        )
    
    def forward(self, x):
        # 重载forward函数，从输入到输出
        return self.dnn(x)

model = Model()  # 初始化模型
creterion = nn.CrossEntropyLoss()  # 选用交叉熵损失函数
optimizer = optim.Adam(model.parameters(), lr=0.001)  # 选用Adam优化器，传入模型参数，设置学习率
for epoch in range(50):  # 50个epoch
    for data, label in dataloader:  # DataLoader是一个可迭代对象
        optimizer.zero_grad()  # 待优化参数梯度清空
        prob = model(data)  # 执行一次前向传播，计算预测结果
        loss = creterion(prob, label)  # 评估模型损失
        loss.backward()  # 损失反向传播，完成对待优化参数的梯度求解
        optimizer.step()  # 参数更新
    if (epoch + 1) % 5 == 0:  # 每隔5个epoch打印当前模型训练效果
        with torch.no_grad():
            train_prob = model(X_train_tensor)
            train_pred = train_prob.argmax(dim=1)
            acc_train = (train_pred==y_train_tensor).float().mean()
            test_prob = model(X_test_tensor)
            test_pred = test_prob.argmax(dim=1)
            acc_test = (test_pred==y_test_tensor).float().mean()
            print(f"epoch: {epoch}, train_accuracy: {acc_train}, test_accuracy: {acc_test} !")
### 输出
epoch: 4, train_accuracy: 0.8507795333862305, test_accuracy: 0.8577777743339539 !
epoch: 9, train_accuracy: 0.948775053024292, test_accuracy: 0.9200000166893005 !
epoch: 14, train_accuracy: 0.9717891812324524, test_accuracy: 0.9444444179534912 !
epoch: 19, train_accuracy: 0.9799554347991943, test_accuracy: 0.9577777981758118 !
epoch: 24, train_accuracy: 0.9866369962692261, test_accuracy: 0.9644444584846497 !
epoch: 29, train_accuracy: 0.9925761222839355, test_accuracy: 0.9644444584846497 !
epoch: 34, train_accuracy: 0.9925761222839355, test_accuracy: 0.9644444584846497 !
epoch: 39, train_accuracy: 0.9962880611419678, test_accuracy: 0.9666666388511658 !
epoch: 44, train_accuracy: 0.9970304369926453, test_accuracy: 0.9711111187934875 !
epoch: 49, train_accuracy: 0.9970304369926453, test_accuracy: 0.9711111187934875 !