logistic regression 实现

 使用ucl中红酒的训练集合，用梯度下降法实现优化

import pandas as pd
import numpy as np
from sklearn import preprocessing
import datetime

class logistic_classfier(object):
    def __init__(self):
        self.col=0  #特征数量
        self.ita=1e-2#学习率
        self.label=[]
        self.theta=[]
    def one_vs_all(self,label):
        #转换标签为[1,0,0,0]之类的并保存在self.label中
        groups=np.unique(label)
        num_label=len(groups)
        label_matrix=np.array([0 for i in range(num_label)])+[[0]for i in range(len(label))] #标签矩阵,使用了numpy数组的broadcast
        #可以用np.expand_dims(a,0).repeat(3,axis=0)来进行复制
        for i in range(len(label)):
            label_matrix[i][label[i]-1]=1
        self.label=label_matrix

    def sigmoid(self,theta,b,x):
        return np.exp(np.dot(theta,x)+b)/(1+np.exp(np.dot(theta,x)+b)) #sigmoid函数

    def cost_function(self,theta,b,feature,label):
        sum=0
        for i in r