python练习-NBA比赛结果预测

# -*- coding: utf-8 -*- """ Created on Sun Jun 4 16:19:16 2017 @author: dc """ import pandas as pd import math import numpy as np from sklearn import linear_model import csv init_elo = 1600 team_elos = {} #trainteamstat = {} folder = 'data' def PruneData(M_stat, O_stat, T_stat): #这个函数要完成的任务在于将原始读入的诸多队伍的数据经过修剪，使其变为一个以team为索引 #排列的特征数据 #丢弃与球队实力无关的统计量 pruneM = M_stat.drop(['Rk', 'Arena'],axis = 1) pruneO = O_stat.drop(['Rk','G','MP'],axis = 1) pruneT = T_stat.drop(['Rk','G','MP'],axis = 1) #将多个数据通过相同的index：team合并为一个数据 mergeMO = pd.merge(pruneM, pruneO, how = 'left', on = 'Team') newstat = pd.merge(mergeMO, pruneT, how = 'left', on = 'Team') #将team作为index的数据返回 return newstat.set_index('Team', drop = True, append = False) def GetElo(team): # try: return team_elos[team] except: team_elos[team] = init_elo return team_elos[team] def CalcElo(winteam, loseteam): # winteam, loseteam的输入应为字符串 # 给出当前两个队伍的elo分数 R1 = GetElo(winteam) R2 = GetElo(loseteam) # 计算比赛后的等级分，参考elo计算公式 E1 = 1/(1 + math.pow(10,(R2 - R1)/400)) E2 = 1/(1 + math.pow(10,(R1 - R2)/400)) if R1>=2400: K=16 elif R1<=2100: K=32 else: K=24 R1new = round(R1 + K*(1 - E1)) R2new = round(R2 + K*(0 - E2)) return R1new, R2new def GenerateTrainData(stat, trainresult): #将输入构造为[[team1特征，team2特征]，...[]...] X = [] y = [] for index, rows in trainresult.iterrows(): winteam = rows['WTeam'] loseteam = rows['LTeam'] #获取最初的elo或是每个队伍最初的elo值 winelo = GetElo(winteam) loseelo = GetElo(loseteam) # 给主场比赛的队伍加上100的elo值 if rows['WLoc'] == 'H': winelo = winelo+100 else: loseelo = loseelo+100 # 把elo当为评价每个队伍的第一个特征值 fea_win = [winelo] fea_lose = [loseelo] # 添加我们从basketball reference.com获得的每个队伍的统计信息 for key, value in stat.loc[winteam].iteritems(): fea_win.append(value) for key, value in stat.loc[loseteam].iteritems(): fea_lose.append(value) # 将两支队伍的特征值随机的分配在每场比赛数据的左右两侧 # 并将对应的0/1赋给y值 if np.random.random() > 0.5: X.append(fea_win+fea_lose) y.append(0) else: X.append(fea_lose+fea_win) y.append(1) # 更新team elo分数 win_new_score, lose_new_score = CalcElo(winteam, loseteam) team_elos[winteam] = win_new_score team_elos[loseteam] = lose_new_score #return np.nan_to_num(X),y return np.nan_to_num(X),y def GeneratePredictData(stat,info): X=[] #遍历所有的待预测数据，将数据变换为特征形式 for index, rows in stat.iterrows(): #首先将elo作为第一个特征 team1 = rows['Vteam'] team2 = rows['Hteam'] elo_team1 = GetElo(team1) elo_team2 = GetElo(team2) fea1 = [elo_team1] fea2 = [elo_team2+100] #球队统计信息作为剩余特征 for key, value in info.loc[team1].iteritems(): fea1.append(value) for key, value in info.loc[team2].iteritems(): fea2.append(value) #两队特征拼接 X.append(fea1 + fea2) #nan_to_num有两个作用：1将列表变换为array，2.去除X中的非数字，保证训练器读入不出问题 return np.nan_to_num(X) if __name__ == '__main__': M_stat = pd.read_csv(folder + '/15-16Miscellaneous_Stat.csv') O_stat = pd.read_csv(folder + '/15-16Opponent_Per_Game_Stat.csv') T_stat = pd.read_csv(folder + '/15-16Team_Per_Game_Stat.csv') team_result = pd.read_csv(folder + '/2015-2016_result.csv') teamstat = PruneData(M_stat, O_stat, T_stat) #'''暂时保存数据''' #teamstat.to_csv(folder + '/Train_data.csv') X,y = GenerateTrainData(teamstat, team_result) limodel = linear_model.LogisticRegression() limodel.fit(X,y) #'''预测''' pre_data = pd.read_csv(folder + '/16-17Schedule.csv') pre_X = GeneratePredictData(pre_data, teamstat) pre_y = limodel.predict_proba(pre_X) #'''显式地显示结果''' predictlist = [] for index, rows in pre_data.iterrows(): reslt = [rows['Vteam'], pre_y[index][0], rows['Hteam'], pre_y[index][1]] predictlist.append(reslt) print(predictlist) #'''预测结果保存''' with open(folder+'/prediction of 2016-2017.csv', 'w',newline='') as f: writers = csv.writer(f) writers.writerow(['Visit Team', 'corresponding probability of winning', 'Home Team', 'corresponding probability of winning']) writers.writerows(predictlist)