使用自己的数据利用pytorch搭建全连接神经网络进行回归预测

1、导入库

引入必要的库，包括pytorch、pandas等。

import numpy as np
import pandas as pd
from sklearn.preprocessing import standardscaler
from sklearn.model_selection import train_test_split
from sklearn.metrics import mean_squared_error, mean_absolute_error
from sklearn.datasets import fetch_california_housing

import torch
import torch.nn as nn
import torch.nn.functional as f
from torch.optim import sgd
import torch.utils.data as data
import matplotlib.pyplot as plt
import seaborn as sns

2、数据准备

这里使用sklearn自带的加利福尼亚房价数据，首次运行会下载数据集，建议下载之后，处理成csv格式单独保存，再重新读取。

后续完整代码中，数据也是采用先下载，单独保存之后，再重新读取的方式。

# 导入数据
housedata = fetch_california_housing()  # 首次运行会下载数据集
data_x, data_y = housedata.data, housedata.target  # 读取数据和标签
data_df = pd.dataframe(data=data_x, columns=housedata.feature_names)  # 将数据处理成dataframe格式
data_df['target'] = data_y  # 添加标签列
data_df.to_csv("california_housing.csv")  # 将数据输出为csv文件
housedata_df = pd.read_csv("california_housing.csv")  # 重新读取数据

3、数据拆分

# 切分训练集和测试集
x_train, x_test, y_train, y_test = train_test_split(housedata[:, :-1], housedata[:, -1],test_size=0.3, random_state=42)

4、数据标准化

# 数据标准化处理
scale = standardscaler()
x_train_std = scale.fit_transform(x_train)
x_test_std = scale.transform(x_test)

5、数据转换

# 将数据集转为张量
x_train_t = torch.from_numpy(x_train_std.astype(np.float32))
y_train_t = torch.from_numpy(y_train.astype(np.float32))
x_test_t = torch.from_numpy(x_test_std.astype(np.float32))
y_test_t = torch.from_numpy(y_test.astype(np.float32))

# 将训练数据处理为数据加载器
train_data = data.tensordataset(x_train_t, y_train_t)
test_data = data.tensordataset(x_test_t, y_test_t)
train_loader = data.dataloader(dataset=train_data, batch_size=64, shuffle=true, num_workers=1)

6、模型搭建

# 搭建全连接神经网络回归
class fnn_regression(nn.module):
    def __init__(self):
        super(fnn_regression, self).__init__()
        # 第一个隐含层
        self.hidden1 = nn.linear(in_features=8, out_features=100, bias=true)
        # 第二个隐含层
        self.hidden2 = nn.linear(100, 100)
        # 第三个隐含层
        self.hidden3 = nn.linear(100, 50)
        # 回归预测层
        self.predict = nn.linear(50, 1)

    # 定义网络前向传播路径
    def forward(self, x):
        x = f.relu(self.hidden1(x))
        x = f.relu(self.hidden2(x))
        x = f.relu(self.hidden3(x))
        output = self.predict(x)
        # 输出一个一维向量
        return output[:, 0]

7、模型训练

# 定义优化器
optimizer = torch.optim.sgd(testnet.parameters(), lr=0.01)
loss_func = nn.mseloss()  # 均方根误差损失函数
train_loss_all = []

# 对模型迭代训练，总共epoch轮
for epoch in range(30):
    train_loss = 0
    train_num = 0
    # 对训练数据的加载器进行迭代计算
    for step, (b_x, b_y) in enumerate(train_loader):
        output = testnet(b_x)  # mlp在训练batch上的输出
        loss = loss_func(output, b_y)  # 均方根损失函数
        optimizer.zero_grad()  # 每次迭代梯度初始化0
        loss.backward()  # 反向传播，计算梯度
        optimizer.step()  # 使用梯度进行优化
        train_loss += loss.item() * b_x.size(0)
        train_num += b_x.size(0)
    train_loss_all.append(train_loss / train_num)

8、模型预测

y_pre = testnet(x_test_t)
y_pre = y_pre.data.numpy()
mae = mean_absolute_error(y_test, y_pre)
print('在测试集上的绝对值误差为:', mae)

9、完整代码

# -*- coding: utf-8 -*-
# @time : 2023/8/11 15:58
# @author : huangjian
# @email : huangjian013@126.com
# @file : fnn_demo.py

import numpy as np
import pandas as pd
from sklearn.preprocessing import standardscaler
from sklearn.model_selection import train_test_split
from sklearn.metrics import mean_squared_error, mean_absolute_error
from sklearn.datasets import fetch_california_housing

import torch
import torch.nn as nn
import torch.nn.functional as f
from torch.optim import sgd
import torch.utils.data as data
from torchsummary import summary
from torchviz import make_dot
import matplotlib.pyplot as plt
import seaborn as sns


# 搭建全连接神经网络回归
class fnn_regression(nn.module):
    def __init__(self):
        super(fnn_regression, self).__init__()
        # 第一个隐含层
        self.hidden1 = nn.linear(in_features=8, out_features=100, bias=true)
        # 第二个隐含层
        self.hidden2 = nn.linear(100, 100)
        # 第三个隐含层
        self.hidden3 = nn.linear(100, 50)
        # 回归预测层
        self.predict = nn.linear(50, 1)

    # 定义网络前向传播路径
    def forward(self, x):
        x = f.relu(self.hidden1(x))
        x = f.relu(self.hidden2(x))
        x = f.relu(self.hidden3(x))
        output = self.predict(x)
        # 输出一个一维向量
        return output[:, 0]


# 导入数据
housedata_df = pd.read_csv("california_housing.csv")
housedata = housedata_df.values
# 切分训练集和测试集
x_train, x_test, y_train, y_test = train_test_split(housedata[:, :-1], housedata[:, -1],test_size=0.3, random_state=42)

# 数据标准化处理
scale = standardscaler()
x_train_std = scale.fit_transform(x_train)
x_test_std = scale.transform(x_test)

# 将训练数据转为数据表
datacor = np.corrcoef(housedata_df.values, rowvar=0)
datacor = pd.dataframe(data=datacor, columns=housedata_df.columns, index=housedata_df.columns)
plt.figure(figsize=(8, 6))
ax = sns.heatmap(datacor, square=true, annot=true, fmt='.3f', linewidths=.5, cmap='ylgnbu',
                 cbar_kws={'fraction': 0.046, 'pad': 0.03})
plt.show()

# 将数据集转为张量
x_train_t = torch.from_numpy(x_train_std.astype(np.float32))
y_train_t = torch.from_numpy(y_train.astype(np.float32))
x_test_t = torch.from_numpy(x_test_std.astype(np.float32))
y_test_t = torch.from_numpy(y_test.astype(np.float32))

# 将训练数据处理为数据加载器
train_data = data.tensordataset(x_train_t, y_train_t)
test_data = data.tensordataset(x_test_t, y_test_t)
train_loader = data.dataloader(dataset=train_data, batch_size=64, shuffle=true, num_workers=1)

# 输出网络结构
testnet = fnn_regression()
summary(testnet, input_size=(1, 8))  # 表示1个样本，每个样本有8个特征

# 输出网络结构
testnet = fnn_regression()
x = torch.randn(1, 8).requires_grad_(true)
y = testnet(x)
mymlp_vis = make_dot(y, params=dict(list(testnet.named_parameters()) + [('x', x)]))

# 定义优化器
optimizer = torch.optim.sgd(testnet.parameters(), lr=0.01)
loss_func = nn.mseloss()  # 均方根误差损失函数
train_loss_all = []

# 对模型迭代训练，总共epoch轮
for epoch in range(30):
    train_loss = 0
    train_num = 0
    # 对训练数据的加载器进行迭代计算
    for step, (b_x, b_y) in enumerate(train_loader):
        output = testnet(b_x)  # mlp在训练batch上的输出
        loss = loss_func(output, b_y)  # 均方根损失函数
        optimizer.zero_grad()  # 每次迭代梯度初始化0
        loss.backward()  # 反向传播，计算梯度
        optimizer.step()  # 使用梯度进行优化
        train_loss += loss.item() * b_x.size(0)
        train_num += b_x.size(0)
    train_loss_all.append(train_loss / train_num)

# 可视化训练损失函数的变换情况
plt.figure(figsize=(8, 6))
plt.plot(train_loss_all, 'ro-', label='train loss')
plt.legend()
plt.grid()
plt.xlabel('epoch')
plt.ylabel('loss')
plt.show()

y_pre = testnet(x_test_t)
y_pre = y_pre.data.numpy()
mae = mean_absolute_error(y_test, y_pre)
print('在测试集上的绝对值误差为:', mae)

# 可视化测试数据的拟合情况
index = np.argsort(y_test)
plt.figure(figsize=(8, 6))
plt.plot(np.arange(len(y_test)), y_test[index], 'r', label='original y')
plt.scatter(np.arange(len(y_pre)), y_pre[index], s=3, c='b', label='prediction')
plt.legend(loc='upper left')
plt.grid()
plt.xlabel('index')
plt.ylabel('y')
plt.show()