朴素贝叶斯算法c++实现

以下是朴素贝叶斯算法的C++实现示例：

#include <iostream>
#include <fstream>
#include <sstream>
#include <vector>
#include <map>
#include <cmath>

using namespace std;

class NaiveBayes {
private:
    vector<vector<double>> trainData; // 训练集数据
    vector<int> trainLabel; // 训练集标签
    int featureSize; // 特征数量
    int classSize; // 类别数量
    map<int, int> classCnt; // 每个类别的样本数量
    map<int, vector<double>> classMean; // 每个类别的特征均值
    map<int, vector<double>> classVar; // 每个类别的特征方差
    map<int, double> classPrior; // 每个类别的先验概率

public:
    NaiveBayes(int featureSize, int classSize) {
        this->featureSize = featureSize;
        this->classSize = classSize;
        trainData.resize(classSize);
        for (int i = 0; i < classSize; i++) {
            trainData[i].resize(featureSize);
        }
    }

    void fit(vector<vector<double>>& X, vector<int>& y) {
        int n = X.size();
        if (n != (int)y.size()) {
            cout << "Error: size of X and y not match!\n";
            return;
        }
        for (int i = 0; i < n; i++) {
            int c = y[i];
            classCnt[c]++;
            for (int j = 0; j < featureSize; j++) {
                trainData[c][j] += X[i][j];
            }
        }
        for (int i = 0; i < classSize; i++) {
            classPrior[i] = (double)classCnt[i] / n;
            for (int j = 0; j < featureSize; j++) {
                classMean[i].push_back(trainData[i][j] / classCnt[i]);
                classVar[i].push_back(0);
            }
        }
        for (int i = 0; i < n; i++) {
            int c = y[i];
            for (int j = 0; j < featureSize; j++) {
                classVar[c][j] += pow(X[i][j] - classMean[c][j], 2);
            }
        }
        for (int i = 0; i < classSize; i++) {
            for (int j = 0; j < featureSize; j++) {
                classVar[i][j] /= classCnt[i];
            }
        }
    }

    int predict(vector<double>& x) {
        int maxC = 0;
        double maxP = -1;
        for (int i = 0; i < classSize; i++) {
            double p = classPrior[i];
            for (int j = 0; j < featureSize; j++) {
                p *= 1 / sqrt(2 * M_PI * classVar[i][j]) * exp(-pow(x[j] - classMean[i][j], 2) / (2 * classVar[i][j]));
            }
            if (p > maxP) {
                maxP = p;
                maxC = i;
            }
        }
        return maxC;
    }
};

int main() {
    // 读取数据
    vector<vector<double>> X;
    vector<int> y;
    ifstream fin("data.txt");
    string line;
    while (getline(fin, line)) {
        istringstream ss(line);
        double val;
        vector<double> vec;
        while (ss >> val) {
            vec.push_back(val);
        }
        X.push_back(vec);
        y.push_back(vec.back());
    }

    // 模型训练
    int featureSize = X[0].size() - 1;
    int classSize = *max_element(y.begin(), y.end()) + 1;
    NaiveBayes clf(featureSize, classSize);
    clf.fit(X, y);

    // 模型预测
    vector<double> x(featureSize);
    x[0] = 5.1;
    x[1] = 3.5;
    x[2] = 1.4;
    x[3] = 0.2;
    int pred = clf.predict(x);
    cout << "Predicted class: " << pred << endl;

    return 0;
}

在该实现中，我们首先定义了一个 NaiveBayes 类，里面包含了模型的训练和预测方法。在训练方法 fit 中，我们统计了每个类别的样本数量、特征均值和特征方差，以及每个类别的先验概率。在预测方法 predict 中，我们计算了每个类别的后验概率，并选择后验概率最大的类别作为预测结果。

该实现中的数据集格式为每行一个样本，以空格分隔各个特征，最后一个特征为标签。如果你想尝试该代码，请先将数据集保存为 data.txt 文件。