NPB の TrueSkill によるレーティング

rate_npb.py

#!/usr/bin/env python
# coding: utf-8

# In[47]:


import itertools
import math
from pathlib import Path


# In[48]:


import matplotlib.pyplot as plt
import numpy as np
import pandas as pd
import trueskill
from trueskill import rate_1vs1


# In[49]:


def win_probability(team1, team2, env=None):
    env = env if env else trueskill.global_env()
    delta_mu = sum(r.mu for r in team1) - sum(r.mu for r in team2)
    sum_sigma = sum(r.sigma ** 2 for r in itertools.chain(team1, team2))
    size = len(team1) + len(team2)
    denom = math.sqrt(size * (env.beta * env.beta) + sum_sigma)
    return env.cdf(delta_mu / denom)


# In[50]:


def expose_and_clip(rating, env=None, minimum=0., maximum=50.):
    env = env if env else trueskill.global_env()
    return min(max(minimum, env.expose(rating)), maximum)


# In[51]:


def sma(src, interval=3):
    if interval > 0 and interval % 2 == 1: pass
    else: raise ValueError('Interval must be positive odd number.')

    margin_num = (interval - 1) // 2
    left_margin = np.full(margin_num, src[0])
    right_margin = np.full(margin_num, src[-1])
    marginal = np.empty(len(src) + 2 * margin_num)
    marginal[:margin_num] = left_margin
    marginal[margin_num:-margin_num] = src[:]
    marginal[-margin_num:] = right_margin

    kernel = np.ones(interval) / interval
    dst = np.convolve(marginal, kernel, mode='valid')
    
    return dst


# In[52]:


def pad_by_nan(teams_transitions):
    games_nums = []
    for teams_transition in teams_transitions.values():
        games_nums.append(len(teams_transition))
    max_game = max(games_nums)
    for team in teams_transitions:
        teams_transition = teams_transitions[team]
        games_dif = max_game - len(teams_transition)
        if games_dif > 0:
            nans = np.full(games_dif, np.nan)
            teams_transitions[team] = np.concatenate((teams_transition,nans,))


# In[53]:


year = 2019


# In[54]:


teams = [
    'Giants', 'Swallows', 'BayStars', 'Dragons', 'Tigers', 'Carp',
    'Fighters', 'Eagles', 'Lions', 'Marines', 'Buffaloes', 'Hawks',]


# In[55]:


teams_colors = {
    'Giants': ('#FF7820','#1E191A','#F79822',),
    'Swallows': ('#073180','#111C3C','#FFFFFF',),
    'BayStars': ('#0052CD','#024A92','#E1F3FD',),
    'Dragons': ('#003595','#042E83','#FFFFFF',),
    'Tigers': ('#FFE100','#000000','#FFD600',),
    'Carp': ('#E50012','#E60012','#FFFFFF',),
    'Fighters': ('#00508F','#000000','#00508F',),
    'Eagles': ('#870010','#86010F','#F5A902',),
    'Lions': ('#00214B','#102960','#FFFFFF',),
    'Marines': ('#000000','#000000','#FFFFFF',),
    'Buffaloes': ('#202030','#010119','#BBAA31',),
    'Hawks': ('#FBC700','#000000','#FCC800',),
}


# In[56]:


is_central = {
    'Giants': True,
    'Swallows': True,
    'BayStars': True,
    'Dragons': True,
    'Tigers': True,
    'Carp': True,
    'Fighters': False,
    'Eagles': False,
    'Lions': False,
    'Marines': False,
    'Buffaloes': False,
    'Hawks': False,
}


# In[57]:


draw_csv = Path('./data/draw.csv')
draw_data = pd.read_csv(draw_csv).set_index('year').loc[year - 3:year - 1].sum()
draw_probability = draw_data['draw'] / draw_data['game']


# In[58]:


env = trueskill.TrueSkill(draw_probability=draw_probability)


# In[59]:


ratings = {}
for team in teams:
    ratings[team] = env.create_rating()


# In[60]:


ratix_transition = {}
mu_transition = {}
sigma_transition = {}


# In[61]:


for team in teams:
    ratix_transition[team] = [expose_and_clip(ratings[team], env=env),]
    mu_transition[team] = [ratings[team].mu,]
    sigma_transition[team] = [ratings[team].sigma,]


# In[62]:


games_dir = Path('./data/games/')
games_csv = games_dir / f'npb_data_{year}.csv'
games_df = pd.read_csv(games_csv)


# In[63]:


for row_tuple in games_df.iterrows():
    row = row_tuple[1]
    
    home = row['home']
    visitor = row['visitor']
    hscore = int(row['hscore'])
    vscore = int(row['vscore'])
    
    card = (ratings[home],),(ratings[visitor],),
    if hscore > vscore:
        ranks = (0,1,)
    elif vscore > hscore:
        ranks = (1,0,)
    else:
        ranks = (0,0,)
    (ratings[home],),(ratings[visitor],), = env.rate(card, ranks=ranks)
    
    ratix_transition[home].append(expose_and_clip(ratings[home], env=env),)
    ratix_transition[visitor].append(expose_and_clip(ratings[visitor], env=env),)
    
    mu_transition[home].append(ratings[home].mu,)
    mu_transition[visitor].append(ratings[visitor].mu,)
    
    sigma_transition[home].append(ratings[home].sigma,)
    sigma_transition[visitor].append(ratings[visitor].sigma,)


# In[64]:


for team in teams:
    ratix_transition[team].reverse()
    mu_transition[team].reverse()
    sigma_transition[team].reverse()


# In[65]:


plt.figure(0)
plt.title('NPB Rating', fontsize=16)
for team in teams:
    x = -np.arange(len(ratix_transition[team]))
    y = sma(ratix_transition[team], 5)
    plt.plot(x, y, label=team,
            marker='.', ms=10, markevery=10,
            c=teams_colors[team][0],
            mec=teams_colors[team][1],
            mfc=teams_colors[team][2],
            ls='-' if is_central[team] else '--')
plt.xlabel('Recent Games', fontsize=14)
plt.ylabel('TrueSkill', fontsize=14)
# plt.ylim(17, 25)
plt.legend(bbox_to_anchor=(1.05, 1), loc='upper left', borderaxespad=0)
plt.savefig('./npb.svg')


# In[66]:


pad_by_nan(ratix_transition)
pad_by_nan(mu_transition)
pad_by_nan(sigma_transition)


# In[67]:


result_dir = Path('./result/')
result_dir.mkdir(parents=True, exist_ok=True)

ratix_csv = result_dir / f'./ratix_{year}.csv'
mu_csv = result_dir / f'./mu_{year}.csv'
sigma_csv = result_dir / f'sigma_{year}.csv'

ratix_df = pd.DataFrame(ratix_transition)
ratix_df.index = -ratix_df.index
ratix_df.reset_index().to_csv(ratix_csv, index=False)

mu_df = pd.DataFrame(mu_transition)
mu_df.index = -mu_df.index
mu_df.reset_index().to_csv(mu_csv, index=False)

sigma_df = pd.DataFrame(sigma_transition)
sigma_df.index = -sigma_df.index
sigma_df.reset_index().to_csv(sigma_csv, index=False)

ratix_df


# In[68]:


wp_columns = {}
for lose_team in teams:
    lose_rating = ratings[lose_team]
    wp_cells = {}
    for win_team in teams:
        win_rating = ratings[win_team]
        wp_cells[win_team] = win_probability((win_rating,), (lose_rating,), env=env)
    wp_columns[lose_team] = wp_cells


# In[69]:


wp_df = pd.DataFrame(wp_columns)
wp_df.index.name = 'win'
wp_df.columns.name = 'lose'
wp_df