有约束条件的加权最小二乘的实现

@[TOC] 有约束条件的加权最小二乘的实现

加权最小二乘原理

详细见网址：加权最小二乘法

Python实现

import numpy as np
import pandas as pd
from scipy.optimize import minimize


class Constrained_regression:
    def __init__(self, weight = None, intercept = True):
        """
        默认选择最小二乘
        """
        self.weight = weight
        self.intercept = intercept
    
    
    def weight_error(self, B, X_train, y_train):
        # 加权最小二乘的损失函数
        # 也是目标函数
        # B表示beta，为待优化参数

        data1 = np.dot(X_train, B)
        data2 = y_train - data1
        
        if self.weight is None: # 此时为普通最小二乘
            self.weight = np.identity(self.sample_num)
            
        weight_1 = np.linalg.pinv(self.weight)
        temp3 = data2.transpose()
        error1 = np.dot(temp3, weight_1)
        error = np.dot(error1, data2)
        return error
    
    
    def cons(self):
        # 创建约束条件
        if self.intercept == True:
            cons = ({'type': 'eq', 'fun': lambda x: np.sum(x[:-1])- 1})
        else:
            cons = ({'type': 'eq', 'fun': lambda x: np.sum(x)- 1})
            
        bnds = []
        for i in range(self.feature_num):
            bnds.append((0, 1))
            
        if self.intercept == True:
            bnds.append((None, None))   
        bnds = tuple(bnds)
        
        return cons, bnds
    
    
    def fit(self, X, y):
        # 模型训练
        X_train = X
        y_train = y
        sample_num = X_train.shape[0]
        feature_num = X_train.shape[1]
        self.feature_num = feature_num
        self.sample_num = sample_num
        
        # 创造截距项
        if self.intercept == True:
            X_train['inter'] = 1
            
        B_ini = list(np.zeros(feature_num) + 1/feature_num)
        B_ini.append(0.5)
        fun = lambda x: self.weight_error(B=x, X_train=X_train, y_train=y_train)
        cons, bnds = self.cons()
        res = minimize(fun, B_ini, method='SLSQP', constraints=cons, tol=1e-5, bounds = bnds)
        return res.x


if __name__ == "__main__":
    
    data = pd.read_csv('YFDdata.csv')
    X_train = data[['bond', 'Conbond', 'Credebt']]
    y_train = data['YFD']
    model = Constrained_regression()
    conf = model.fit(X_train, y_train)

第一次写CSDN，比较简单，下次会修正