逻辑回归丨对数几率回归算法实现

对线性回归算法的添加关系函数

通过最大似然法得到对数似然函数

为最大化似然函数，函数对求导，沿着梯度方向优化

直接上代码实现

import numpy as np
import pandas as pd

class RegressionBase(object):
    def __init__(self):
        self.bias = None

class logic_Regression(RegressionBase):
    def __init__(self):
        RegressionBase.__init__(self)

    def sigmoid(self,x):
        return 1.0/(1+np.exp(-1*x))

    def fit(self, X, Y, epos=10, eta=0.1, bias=0):
        self.bias = bias
        self.W = np.zeros(X.shape[1]+1)
        for i in range(epos):
            grad_delta = np.zeros(X.shape[1]+1)
            for j in range(X.shape[0]):
                X_ = np.concatenate(([1], X[j,:]))
                grad_delta += Y[j]*X_ - X_*self.sigmoid((self.W*X_).sum())
            grad_delta *= 1.0/X.shape[0]
            self.W += eta*grad_delta

    def _predict(self, Xi):
        return self.sigmoid((Xi*self.W[1:]).sum()+self.W[0]+self.bias)

    def predict(self, X):
        return [int(self._predict(Xi)>0.5) for Xi in X]

if __name__ == '__main__':
    from sklearn import datasets
    from sklearn.model_selection import train_test_split
    from sklearn.metrics import classification_report

    iri = datasets.load_iris()
    X = iri['data'][:100]
    y = iri['target'][:100]
    X_train,X_test,y_train,y_test = train_test_split(X, y, train_size=0.7, test_size=0.3)
    lrg = logic_Regression()
    lrg.fit(X_train, y_train)
    y_pred = lrg.predict(X_test)
    print(classification_report(y_test, y_pred))