run_glicko2_mma.pyx

def calc_glicko2_rank(df: pd.DataFrame, tau:float, rd_0:float, sigma_0:float = 0.06, date_freq:str = '180D') -> pd.DataFrame:
    glicko_scorer = Glicko2(tau=tau)
    fighter_set = set(df['fighter_slog'].unique().tolist() + df['opponent_slog'].unique().tolist())
    for fighter in fighter_set:
        glicko_scorer.add_player(fighter, rd=rd_0, sigma=sigma_0)
    
    for group_idx, (date_idx, date_group) in tqdm(enumerate(df.groupby([pd.Grouper(key='date', freq=date_freq)]))):

        # pre-rating map
        rating_map = {fname: glicko_scorer[fname].rating for fname in fighter_set}
        df.loc[date_group.index, 'pre_glicko_f1'] = date_group['fighter_slog'].map(rating_map)
        df.loc[date_group.index, 'pre_glicko_f2'] = date_group['opponent_slog'].map(rating_map)

        rd_map = {fname: glicko_scorer[fname].rd for fname in fighter_set}
        df.loc[date_group.index, 'pre_glicko_rd_f1'] = date_group['fighter_slog'].map(rd_map)
        df.loc[date_group.index, 'pre_glicko_rd_f2'] = date_group['opponent_slog'].map(rd_map)
        
        # Add fights from each fighter
        for fs_row in date_group.itertuples():
                
            fighter = fs_row.fighter_slog
            opponent = fs_row.opponent_slog
            score = fs_row.win_method_score if fs_row.result != 'loss' else 0.
            glicko_scorer[fighter].add_fight(opponent, score)
            
        
        
        # Get updates for each fighter
        new_ratings = {}
        for fname in fighter_set:
            if len(glicko_scorer[fname].fights) > 0:
                new_ratings[fname] = glicko_scorer.update_ratings(fname)
            else:
                new_ratings[fname] = glicko_scorer.update_ratings_no_fights(fname)
        
        # Apply updates and reset fights
        for fname  in fighter_set:
            glicko_scorer[fname].reset_fights()
            glicko_scorer[fname].update_params(*new_ratings[fname])
    
    return df