elo_football.py

from __future__ import division
import numpy as np
import pandas as pd
import random
from datetime import date


def expected(A, B):
    """
    Calculate expected score of team A in a match against team B
    
    parameters
    -------------------------
    A: float
        Elo rating for team A
    B: float
        Elo rating for team B
    """
    return 1 / (1 + 10 ** ((B - A) / 400))


def elo(old, exp, score, k=32):
    """
    Calculate the new Elo rating for the players in a given team

    parameters
    --------------------------
    old: float
        The previous Elo rating
    exp: float
        The expected probability of winning for this match
    score: int
        The actual score for this match (1 for win, 0 for loss)
    k: int
        The k-factor for Elo (default: 32)
    """
    return old + k * (score - exp)

# read in the latest data
filename = 'C:/Users/aluz/Desktop/elo_football/elo_football.csv'
df = pd.read_csv(filename)

# have the teams already been selected? 
x = raw_input('Have the teams already been selected (y or n)? >')

# if they have...
if x == 'y':
    # input the names of the players who play for team 1
    team_1_str = raw_input('Players in team 1 (separate with a comma) >')
    # and convert that string to a list where each item is a player of team 1
    team_1 = team_1_str.split(',')
    # input the names of the players who play for team 2
    team_2_str = raw_input('Players in team 2 (separate with a comma) >')
    # and convert that string to a list where each item is a player of team 2
    team_2 = team_2_str.split(',')

    # calculate the team ratings by adding up the individual ratings of the players in each team 
    team_1_ratings = df.loc[df.player.isin(team_1), 'rating'].sum()
    team_2_ratings = df.loc[df.player.isin(team_2), 'rating'].sum()

    # calculate the expected probabilities of winning for each team,
    # according to the Elo expected function defined above
    exp_1 = expected(team_1_ratings, team_2_ratings) 
    exp_2 = 1 - exp_1

    # print the probability of winning for team 1, as well as what players are playing in each team
    print 'Probability of team 1 winning = ', exp_1
    print 'Team 1: ', team_1
    print 'Team 2: ', team_2
# if the teams haven't been selected yet...
else:
    # input the names of all the players that are available for the game (8 to 12 players)
    available = raw_input('List of players that are playing today (separate with a comma) >')
    # and convert that string to a list where each item is a player
    available_list = available.split(',')
    # calculate the number of players in team 1
    # if the length of the list is an even number, n will be equal to the number of players in each team
    # if the length of the list is an odd number, n will be equal to the number of players in team 1
    # team 2 will have len(available_list) - n players
    # e.g. if there are 9 players, int(9 / 2) = 4, so team 1 will have 4 players and team 2 will have 5 players
    n = int(len(available_list) / 2)
    
    # initialize an empty list of players for team 1
    team_1 = []

    # initialize exp and iterations variables
    exp = 0
    iterations = 0

    # repeat the following while the teams are not balanced enough
    # for them to be balanced, the probability of winning for team 1
    # has to be between 40 and 60%
    while (exp <= 0.40)  | (exp >= 0.60):
        # re-create the list of all players every time
        available_list = available.split(',')
        team_1 = []

        # do this as many times as there are players in team 1
        for i in range(n):
            # select a random player from the whole list of players available
            name = random.choice(available_list)
            # add that name to team 1
            team_1.append(name)
            # remove that name from the list to ensure we don't draw it again
            available_list.remove(name)

        # team 2 will be formed by all the players available that are not in team 1
        team_2 = [player for player in available.split(',') if player not in team_1]

        # calculate the ratings for each team, based on this random split
        team_1_ratings = df.loc[df.player.isin(team_1), 'rating'].sum()
        team_2_ratings = df.loc[df.player.isin(team_2), 'rating'].sum()

        # calculate the expected probability of winning for team 1
        # if this is not between 40 and 60%, the loop will run again
        exp = expected(team_1_ratings, team_2_ratings)
        iterations += 1

    # set the final expectations of winning for each team
    exp_1 = exp
    exp_2 = 1 - exp_1

    # print how many iterations it took the while loop to find a balanced match
    # as well as the expected probability of winning for team 1
    # and who's in each team
    print iterations
    print 'Probability of team 1 winning = ', exp_1
    print 'Team 1: ', team_1
    print 'Team 2: ', team_2

# after the game, input the number of the team that won the match
winning_team = raw_input('Who won? 1 or 2? >')

# if team 1 won, set their score to 1 and team 2's score to 0, and viceversa
if winning_team == '1':
    score_1 = 1
    score_2 = 0
else:
    score_1 = 0
    score_2 = 1

# calculate the difference between the current team rating and the new Elo rating
# based on the elo function defined above
diff_1 = elo(team_1_ratings, exp_1, score_1) - team_1_ratings
diff_2 = elo(team_2_ratings, exp_2, score_2) - team_2_ratings

# create a copy of the original table
df_new = df.copy()

# adjust the ratings of each player based on whether they won or lost
# the adjustment is the difference calculated above
df_new.loc[df_new.player.isin(team_1), 'rating'] += diff_1
df_new.loc[df_new.player.isin(team_2), 'rating'] += diff_2

# overwrite the original file with the new file, which has the adjusted player ratings
df_new.to_csv(filename, index=False)

# read in the historical data
filename_hist = 'C:/Users/aluz/Desktop/elo_football/elo_football_history.csv'
df_hist = pd.read_csv(filename_hist)

# format the date properly
df_hist.date = pd.to_datetime(df_hist.date, format='%d/%m/%Y')

# add the new (most up-to-date) ratings to the historical ones
df_hist_expanded = pd.concat([df_hist, df_new], axis=0)

# add today's date as the date for the new ratings
# we want to add the dates so in the future we will be able to plot the ratings over time
df_hist_expanded.date.fillna(date.today(), inplace=True)

# save the historical data
df_hist_expanded.to_csv(filename_hist, index=False)