multiple_linear_regression.py

def mlr(fpath, fmodel='csv', isdropna=True, droplist, smatrix, n, characteristic_list, mark):
    '''
    author: ky0ha
    title: A classifier by multiple linear regression
    argument: 
        fpath: a string, is the position of data-file
        fmodel: a string, is the model of data-file (default is 'csv')
        isdropna: a bool, drop NAN of data? True is drop, False isn't drop (defalut is True)
        droplist: a list, value of the list is characteristics of data want to del, all characteristics is string
        smatrix: a matrix, shape is (n, 1), to change the value of string to be the number type, sample: 
                [["characteristric1", {"string1": value1}], 
                 ["characteristric2", {"string2": value2}]] meanscharacteristric1's string1 change to value1
        n: a int, is the length of data to be training
        characteristic_list: a list, value in list is string, means the list of the characteristics
        mark: a string, is the mark of data
    return: a array, is the argument matrix, can actually use in estimate fuction
    '''
    #----------------------------------------------------------------------------------------------------------------------

    #方法构建部分
    def average(x, n):       #x为列表，n为列表个数，函数效果为求x的平均值
        total=0
        for i in range(n):
            total+=x[i]
        return total/n

    def getTheta(X_b, y):        #X_b为xi的训练集数值构成的矩阵， y为x对应的y的值构成的向量，函数效果为根据X_b和y求出x的参数矩阵
        theta_best = np.linalg.inv(X_b.T.dot(X_b)).dot(X_b.T).dot(y)
        return theta_best

    #--------------------------------------------------------------------------------------------------------------------

    #引入库
    import pandas as pd
    import numpy as np
    from math import sqrt
    from matplotlib import pyplot as plt

    #打开文件
    # fpath = "C:/Users/25315/Desktop/Churn-Modelling.csv"
    if fmodel=='csv':
        f = pd.read_csv(fpath)
    elif fmodel=='excel':
        f = pd.read_excel(fpath)
    else:
        print("wrong file model, only can be 'csv' or 'excel'!")
        return 0

    #数据清洗部分
    #剔除无关数据、相关性低数据（相关性指数低于0.001）、空数据
    # f.drop(["RowNumber","CustomerId","Surname", "HasCrCard"], axis=1, inplace=True)
    f.drop(droplist, axis=1, inplace=True)
    if isdropna:
        f.dropna(axis=0,how='any',thresh=None,subset=None,inplace=True)
    #剔除读取自带的索引index
    f=f.reset_index(drop=True)
    #字符类型数据替换为数值类型
    for s in smatrix:
        f[s[0]] = f[s[0]].map(s[1]) 
    dic = {"France":1, "Spain":2, "Germany":3}
    f["Geography"] = f["Geography"].map(dic)
    dicf = {"Female":0, "Male":1}
    f["Gender"] = f["Gender"].map(dicf)

    #拆分90%作为训练集
    # n = len(f)
    # n1 = int(len(f)*0.9)

    #数据提取赋值部分
    CL = []
    for i in characteristic_list:
        CL.append(np.array(list(f.loc[:,i]))[:n])
    # CS = f.loc[:,"CreditScore"]
    # GY = f.loc[:,"Geography"]
    # GR = f.loc[:,"Gender"]
    # AGE = f.loc[:,"Age"]
    # TE = f.loc[:,"Tenure"]
    # BL = f.loc[:,"Balance"]
    # NOP = f.loc[:,"NumOfProducts"]
    # IAM = f.loc[:,"IsActiveMember"]
    # ES = f.loc[:,"EstimatedSalary"]
    # EXIT = f.loc[:,"Exited"]

    #训练集数据切分部分
    # x1 = np.array(list(CS))[:n1]
    # x2 = np.array(list(GY))[:n1]
    # x3 = np.array(list(GR))[:n1]
    # x4 = np.array(list(AGE))[:n1]
    # x5 = np.array(list(TE))[:n1]
    # x6 = np.array(list(BL))[:n1]
    # x7 = np.array(list(NOP))[:n1]
    # x8 = np.array(list(IAM))[:n1]
    # x9 = np.array(list(ES))[:n1]
    # y = np.array(list(EXIT))[:n1]

    #模型构建过程
    #获得x矩阵
    X_b = np.ones((n,1))
    y = np.array(list(f.loc[:,mark]))[:n]
    for i in CL:
        X_b = np.c_[X_b, i]
    #带入x矩阵求得参数矩阵theta_best
    theta_best = getTheta(X_b,y)
    return theta_best

    #----------------------------------------------------------------------------------------
def estimate(theta_best, X, n=1):         #预测模型的函数模型，根据参数矩阵和输入的x矩阵，求得对应的预测结果y的向量
    X=np.c_[np.ones(n),X]
    X=X.T
    theta_best=np.array([theta_best])
    return theta_best.dot(X)
    #测试部分

    #测试集数据切分部分
    x1 = np.array(list(CS))[n1:]
    x2 = np.array(list(GY))[n1:]
    x3 = np.array(list(GR))[n1:]
    x4 = np.array(list(AGE))[n1:]
    x5 = np.array(list(TE))[n1:]
    x6 = np.array(list(BL))[n1:]
    x7 = np.array(list(NOP))[n1:]
    x8 = np.array(list(IAM))[n1:]
    x9 = np.array(list(ES))[n1:]
    y = np.array(list(EXIT))[n1:]

    #待测数据处理
    X_t = np.c_[x1,x2,x3,x4,x5,x6,x7,x8,x9]

    #测试过程
    es = estimate(theta_best,X_t,int(n*0.1))[0]

    #测试结果处理统计
    xe = list(map(round, es.tolist()))
    re = y-xe
    right = re.tolist().count(0)
    wrong = int(n*0.1)-right
    rp = right/int(n*0.1)

    #测试结果展示
    print("数据集共有{}个数据。模型训练集共有{}个数据，占总数据集的90%。测试集共有{}个数据，占总数聚集的10%。\n".format(n, int(n*0.9), int(n*0.1)))
    print("测试结果显示：\n")
    print("总测试数据个数：{}\n预测正确数据个数：{}\n预测错误数据个数：{}\n模型预测准确率：{:.2f}%".format(int(n*0.1), right, wrong, rp))

    #测试结果柱状图展示
    TR = ["right", "wrong"]
    TA = [right, wrong]
    plt.title("the result of test") 
    plt.xlabel("test result") 
    plt.ylabel("amount") 
    for a,b in zip(TR,TA):
        plt.text(a, b+2, b, ha='center', va= 'bottom')
    plt.ylim([0, 1000])
    plt.bar(TR,TA,0.5)
    plt.show()



smatrix = [["Geography", {"France":1, "Spain":2, "Germany":3}], ["Gender", {"Female":0, "Male":1}]]