[代码案例] pytorch快速上手写机器学习

任务背景

给定未来一段时间的温度，使用神经网络预测输出是天气炎热，温暖，凉爽，偏冷，寒冷
输入是未来 20天内的气温数据，输出标签是 0,1,2,3,4

代码

"""
    @Author : 琛歌很无聊
    @Description: 使用 Pytorch的简单demo：给定一段时间的温度，输出是天气炎热，温暖，凉爽，偏冷，寒冷
                  输入是未来 20天内的气温数据，输出标签是 0,1,2,3,4
"""
import numpy as np
import torch
import torch.utils.data as Data
import torch.nn as nn
import torch.optim as optim
import matplotlib.pyplot as plt


def getlabel(X):  # 获取样本对应的标签
    Y = np.zeros(len(X))
    for i in range(len(X)):
        if np.mean(X[i]) > 30:
            Y[i] = 0  # 平均温度大于30时，标签设置为炎热（0）
        elif np.mean(X[i]) > 20:
            Y[i] = 1
        elif np.mean(X[i]) > 10:
            Y[i] = 2
        elif np.mean(X[i]) > 0:
            Y[i] = 3
        else:
            Y[i] = 4

    return Y


class MyDataset(Data.Dataset):  # 继承父类
    def __init__(self, train_x, train_y):
        self.train_x = train_x
        self.train_y = train_y

    def __getitem__(self, item):
        return self.train_x[item], self.train_y[item]

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


class MyNeuralNet(nn.Module):  # 继承父类
    def __init__(self, input_dim, output_dim):
        super(MyNeuralNet, self).__init__()
        self.activation = nn.ReLU()  # 激活函数
        # 定义网络
        self.n1 = nn.Linear(input_dim, 16)
        self.n2 = nn.Linear(16, 8)
        self.n3 = nn.Linear(8, output_dim)

    def forward(self, x):
        '''
        :param x: 输入矩阵
        :return: 网络输出值
        '''
        y = self.n1(x)
        y = self.activation(y)
        y = self.n2(y)
        y = self.n3(y)
        return y


'''
1.获取训练集和测试集
'''
train_X = np.random.uniform(33, 42, (100, 20))  # 随机生成温度值域[33,42)，大小为100*20的数组，100是样本数，20是维度数
train_X = np.append(train_X, np.random.uniform(19, 35, (100, 20)), axis=0)  # np.append是数组拼接
train_X = np.append(train_X, np.random.uniform(5, 24, (100, 20)), axis=0)  # np.append是数组拼接
train_X = np.append(train_X, np.random.uniform(-2, 11, (100, 20)), axis=0)  # np.append是数组拼接
train_X = np.append(train_X, np.random.uniform(-15, 3, (100, 20)), axis=0)  # np.append是数组拼接
train_Y = getlabel(train_X)
# print(train_Y)
print("训练集输入的大小为：", np.shape(train_X))

test_X = np.random.uniform(30, 40, (1, 20))
test_Y=getlabel(test_X)

'''
2.加载数据集和加载模型
'''
train_X = torch.Tensor(train_X)  # 将array转成Tensor,float类型
train_Y = torch.LongTensor(train_Y)  # LongTensor中数字是整数
test_X = torch.Tensor(test_X)  # 将array转成Tensor,float类型

train_dataset = MyDataset(train_x=train_X, train_y=train_Y)
train_loader = Data.DataLoader(train_dataset, batch_size=16, shuffle=True)  # batch_size控制一批样本的数量，shuffle控制是否打乱样本取样

model = MyNeuralNet(input_dim=20, output_dim=5)  # 定义模型网络
loss_fun = nn.CrossEntropyLoss()  # 定义损失函数
optimizer = optim.AdamW(model.parameters(), lr=1e-3)  # lr是学习率

'''
3.在模型上训练
'''
max_epoch = 100
loss_list = []
for epoch in range(max_epoch):
    for i, (x, y) in enumerate(train_loader):
        predict_y = model(x)  # 1.做向前计算
        loss = loss_fun(predict_y, y)  # 2.计算损失函数
        optimizer.zero_grad()  # 3.清除网络状态
        loss.backward()  # 4.loss反向传播
        optimizer.step()  # 5.更新网络参数

        # 输出Loss,并存储
        print(loss)
        loss_list.append(loss.item())

'''
4.绘制损失函数曲线图
'''
print(loss_list)
plt.plot(loss_list)
plt.title("loss")
plt.show()

'''
5.使用训练好的模型进行泛化测试
'''
print("测试集是\n", test_X)

predict_Y = model(test_X)
predict_Y = torch.softmax(predict_Y, dim=1)  # 将输出值映射到[0,1]概率分布区间
predict_Y = predict_Y[0].detach().numpy()  # Tensor转数组array
predict_Y = np.round(predict_Y * 100, 3)  # 乘以100为百分比，保留3位小数
print("预测标签概率分布")
print(predict_Y)
print("预测标签值")
print(np.argmax(predict_Y, 0))  # 取最大值的下标为预测值
print("实际标签值")
print(test_Y)

结果展示

测试集是
 tensor([[38.7667, 37.0874, 30.7980, 34.7945, 37.9251, 39.0055, 37.2950, 38.5757,
         30.0081, 36.3308, 38.7851, 39.2595, 35.2415, 38.8545, 35.7583, 32.1818,
         35.3024, 32.5636, 32.8940, 34.4463]])
预测标签概率分布
[99.759  0.241  0.     0.     0.   ]
预测标签值
0
实际标签值
[0.]

代码讲解地址

https://www.bilibili.com/video/BV1KX4y177x8/?spm_id_from=333.999.0.0&vd_source=de24eb60706a87145c55dda9edb79815