示例#1
0
    def __init__(self):
        conn = establish_db_connection('sqlalchemy').connect()
        self.url = "https://kenpom.com/team.php"
        self.url_team = lambda x, y: '%s?team=%s&y=%s' % (self.url, str(x),
                                                          str(y))
        self.df = pd.DataFrame()

        teams = pd.read_sql(
            'SELECT DISTINCT Team FROM kenpom_season_data WHERE Year = 2019 AND Rank <= 60',
            con=conn)
        teams = teams['Team'].tolist()
        skip_teams = pd.read_sql(
            'SELECT DISTINCT Team FROM kenpom_madness_table', con=conn)
        skip_teams = skip_teams['Team'].tolist()
        for team in skip_teams:
            teams.remove(team)

        for team in teams:
            self._start_session(team)
            print "Fetching game results for %s" % team
            self.team = team
            self._fetch_team_results()
            print "Uploading to db..."
            self._upload_data()
            print ""

        print self.df
        self.driver.close()
示例#2
0
def four_factors_algo(team_a, team_b):
    conn = establish_db_connection('sqlalchemy').connect()

    team_a = str(team_a)
    team_b = str(team_b)

    ## fetch data
    cols = {
        'Team': 'team',
        'eFG Perc': 'efg_perc',
        'Turnover Perc': 'tov_perc',
        'Off Rebound Perc': 'orb_perc',
        'FTRate': 'ft_rate',
        'Opp eFG Perc': 'opp_efg_perc',
        'Opp Turnover Perc': 'opp_tov_perc',
        'Opp Off Rebound Perc': 'opp_orb_perc',
        'Opp FTRate': 'opp_ft_rate'
    }
    sql_str = 'SELECT `{}` FROM kenpom_four_factors_data WHERE Year = 2019'.format(
        '`, `'.join(cols.keys()))
    df = pd.read_sql(sql_str, con=conn)
    df.rename(index=str, columns=cols, inplace=True)

    ## clean data
    for col in [col for col in df.columns if col != 'team']:
        df[col] = df[col] / 100

    df = df[df['team'].isin([team_a, team_b])].copy()

    ## run algo
    weights_dict = {'efg': 0.40, 'tov': 0.25, 'orb': 0.20, 'ft_rate': 0.15}

    temp = {}
    for index, row in df.iterrows():
        efg = (row['efg_perc'] - row['opp_efg_perc']) * 100
        tov = (row['tov_perc'] - row['opp_tov_perc']) * 100
        orb = (row['tov_perc'] - row['opp_tov_perc']) * 100
        ft_rate = (row['ft_rate'] - row['opp_ft_rate']) * 100
        temp[row['team']] = {
            'efg': efg,
            'tov': tov,
            'orb': orb,
            'ft_rate': ft_rate
        }

    algo_dict = {}
    print team_a
    for s, w in weights_dict.iteritems():
        algo_dict[s] = (temp[team_a].get(s) - temp[team_b].get(s)) * w

    spread = round((sum(algo_dict.values()) * 2), 3)
    print spread

    return spread
示例#3
0
    def __init__(self, start, end):
        self.conn = establish_db_connection('sqlalchemy').connect()
        self.dates = range_all_dates(start, end)
        self.clean_cols = [
            'game_date', 'game_id', 'away_id', 'away_team', 'home_id',
            'home_team', 'pts_away', 'pts_home', 'pt_diff_away', 'pts_total',
            'win_side', 'win_id'
        ]
        self.clean_scores = pd.DataFrame(columns=self.clean_cols)

        self._fetch_raw_scores()
示例#4
0
 def __init__(self):
     self.conn = establish_db_connection('sqlalchemy').connect()
     self.games = ['', '', '', '', '', '']
     self.stats = [
         'EFG_PCT', 'FTA_RATE', 'TM_TOV_PCT', 'OREB_PCT', 'OPP_EFG_PCT',
         'OPP_FTA_RATE', 'OPP_TOV_PCT', 'OPP_OREB_PCT'
     ]
     self._fetch_stats()
     self._weight_player_stats()
     print self._team[self.stats[:] + ['TEAM_ID']]
     print self._weighted
示例#5
0
    def __init__(self):
        self.conn = establish_db_connection('sqlalchemy').connect()

        self.today = datetime.now().strftime('%Y-%m-%d')
        self.prev_day = (datetime.strptime(self.today, '%Y-%m-%d').date() -
                         timedelta(days=1)).strftime('%Y-%m-%d')
        self.games_list = get_games_list(self.prev_day)

        self._fetch_game_stats()
        self._fetch_final_scores()
        self._fetch_spreads()
        # self._calculate_results()
        self._update_db()
示例#6
0
def linear_algo(team_a, team_b):
    conn = establish_db_connection('sqlalchemy').connect()

    ## fetch data
    cols = {
        'Team': 'team',
        'eFG Perc': 'efg_perc',
        'Turnover Perc': 'tov_perc',
        'Off Rebound Perc': 'orb_perc',
        'FTRate': 'ft_rate',
        'Opp eFG Perc': 'opp_efg_perc',
        'Opp Turnover Perc': 'opp_tov_perc',
        'Opp Off Rebound Perc': 'opp_orb_perc',
        'Opp FTRate': 'opp_ft_rate'
    }
    sql_str = 'SELECT `{}` FROM kenpom_four_factors_data WHERE Year = 2019'.format(
        '`, `'.join(cols.keys()))
    df = pd.read_sql(sql_str, con=conn)
    df.rename(index=str, columns=cols, inplace=True)
    ## adjust percentages for algorithm
    for col in [
            x for x in df.columns
            if x not in ['team', 'opp_ft_rate', 'ft_rate']
    ]:
        df[col] = df[col] / 100

    ## run algo
    weights_dict = {
        'efg_perc': 101.54,
        'tov_perc': -0.829,
        'orb_perc': 0.063,
        'ft_rate': 7.57,
        'opp_efg_perc': -92.41,
        'opp_tov_perc': 1.0428,
        'opp_orb_perc': -0.0123,
        'opp_ft_rate': -13.095
    }
    spreads = {}
    for team in [team_a, team_b]:
        data = df[df['team'] == team].copy()
        spread = 0
        for s, w in weights_dict.iteritems():
            spread += data[s].max() * w
        spreads[team] = spread

    spread = spreads[team_a] - spreads[team_b]
    print spread

    return None
示例#7
0
    def __init__(self, gamedate=None):
        self.conn = establish_db_connection('sqlalchemy').connect()
        if gamedate is not None:
            self.gamedate = gamedate
        else:
            self.gamedate = datetime.now().date().strftime('%Y-%m-%d')

        self.games_df = assets.games_daily(self.gamedate)
        self._fetch_agg_stats()
        self._determine_b2b()
        self._build_predictions()
        self._merge_with_spreads()
        self._clean_merged_data()

        self.merged_df.to_sql('daily_picks',
                              con=self.conn,
                              if_exists='replace',
                              index=False)
        self.merged_df.to_sql('historical_picks',
                              con=self.conn,
                              if_exists='append',
                              index=False)
示例#8
0
 def __init__(self, stats):
     self.conn = establish_db_connection('sqlalchemy').connect()
     self.stats = stats
示例#9
0
 def _upload_data(self):
     conn = establish_db_connection('sqlalchemy').connect()
     self.df.to_sql('kenpom_madness_table',
                    con=conn,
                    if_exists='append',
                    index=None)
示例#10
0
 def __init__(self):
     self.conn = establish_db_connection('sqlalchemy').connect()
示例#11
0
import dash
import dash_core_components as dcc
import dash_html_components as html
import plotly.graph_objs as go
from dash.dependencies import Input, Output, State, Event
import dash_table
import styling
import pandas as pd
import madness_helper as helper
from nba_utilities.db_connection_manager import establish_db_connection

app = dash.Dash()
app.title = "Madness"
app.css.append_css({'external_url':"https://stackpath.bootstrapcdn.com/bootstrap/4.2.1/css/bootstrap.min.css"})
conn = establish_db_connection('sqlalchemy').connect()

teams = pd.read_sql('SELECT DISTINCT team FROM kenpom_four_factors_data', con=conn)
teams = [ row['team'] for index, row in teams.iterrows() ]
teams = [ {'label': team, 'value': team} for team in sorted(teams) ]

def serve_layout():
    return html.Div(children = [
        html.Div(className="app-header", children = [
            html.H1('Alge-bracket', className="app-header--title", style=styling.portal_header_txt)], style = styling.portal_header_bgrd),
            dcc.Tabs(id="tabs", children=[

                #### Head 2 Head Predictor Tab ####
                dcc.Tab(label='H2H Predictor', children=[
                    html.H1(children='Matchup Predictor', style=styling.tab_header),
                    html.Div(style={'padding':10}),
                    html.Div(className='row', children=[