elo_features.py

#!/usr/bin/env python3
# -*- coding: utf-8 -*-
import pandas as pd
import numpy as np

def compute_elo_rankings(data):
    """
    Given the list on matches in chronological order, for each match, computes 
    the elo ranking of the 2 players at the beginning of the match
    
    """
    print("Elo rankings computing...")
    players=list(pd.Series(list(data.Winner)+list(data.Loser)).value_counts().index)
    elo=pd.Series(np.ones(len(players))*1500,index=players)
    ranking_elo=[(1500,1500)]
    for i in range(1,len(data)):
        w=data.iloc[i-1,:].Winner
        l=data.iloc[i-1,:].Loser
        elow=elo[w]
        elol=elo[l]
        pwin=1 / (1 + 10 ** ((elol - elow) / 400))    
        K_win=32
        K_los=32
        new_elow=elow+K_win*(1-pwin)
        new_elol=elol-K_los*(1-pwin)
        elo[w]=new_elow
        elo[l]=new_elol
        ranking_elo.append((elo[data.iloc[i,:].Winner],elo[data.iloc[i,:].Loser])) 
        if i%5000==0:
            print(str(i)+" matches computed...")
    ranking_elo=pd.DataFrame(ranking_elo,columns=["elo_winner","elo_loser"])    
    ranking_elo["proba_elo"]=1 / (1 + 10 ** ((ranking_elo["elo_loser"] - ranking_elo["elo_winner"]) / 400))   
    return ranking_elo