【競馬】いちばん強いのはどの子？～TrueSkillによる競走馬レーティング～

import pandas as pd
import numpy as np
import trueskill

mu = 25.
sigma = mu / 3.
beta = sigma / 2.
tau = sigma / 100.
draw_probability = 0.001
backend = None
env = trueskill.TrueSkill(
   mu=mu, sigma=sigma, beta=beta, tau=tau,
   draw_probability=draw_probability, backend=backend)

race_list = list(df_rating.loc[:,'RaceID'].unique())
horse_list = list(df_rating.loc[:,'HorseID'].unique())

rate_dict = {k:env.create_rating() for k in horse_list}
rate_dict = {k:(v,v) for k,v in rate_dict.items()}

rate_before_ = []
race_after_ = []
for race in race_list:
    df_race = df_rating[df_rating.loc[:,'RaceID'] == race]
    rate_before = [env.expose(rate_dict[el[2]][0]) for el in list(df_race.values)]
    teams = [(rate_dict[el[2]][0],) for el in list(df_race.values)]
    teams = env.rate(teams, ranks=list(range(len(df_race))))
    rate_after = [env.expose(t[0]) for t in teams]
    for i, el in enumerate(list(df_race.values)):
        rate_dict[el[2]] = (teams[i][0], max(rate_dict[el[2]][0], teams[i][0]))
    rate_before_ += rate_before
    race_after_ += rate_after

df_rating.loc[:,'レース前レート'] = rate_before_
df_rating.loc[:,'レース後レート'] = race_after_

display(df_rating)

df_rating=df_rating[['HorseID','馬名','Date','レース後レート']]

df_rating.sort_values('レース後レート', ascending=False).drop_duplicates(subset='HorseID').head(20)

df_rating2020=df_rating.query("HorseID.str.startswith('2020')")
df_rating2020.sort_values('レース後レート', ascending=False).drop_duplicates(subset='HorseID').head(20)