def proj_vs_actual(start_date, end_date):
compare = None
days = (end_date - start_date).days
for offset in range(days + 1):
cur_date = start_date + dt.timedelta(days = offset)
try:
stats = pd.read_csv(format_fpath('stat', cur_date))
except:
continue
stats.loc[:,'Starters'] = stats.Starters.apply(lambda x: format_name(x))
lineups = pd.read_csv(format_fpath('line',cur_date))
lineups.loc[:,'Name'] = lineups.Name.apply(lambda x: format_name(x))
combo = lineups.join(stats.set_index('Starters').FP.rename('actual'), on = 'Name').sort_values('Name').set_index('Name')
combo['date'] = cur_date
mapper = player_team_map()
listed = combo.index.to_series().apply(lambda x: x in mapper.index)
combo['PTeam'] = 'UNK'
combo.loc[listed, 'PTeam'] = combo[listed].index.to_series().apply(lambda x: mapper.loc[x]).values
combo['Loc'] = combo.apply(lambda x: 'Home' if x.PTeam == x.Team else 'Away', axis = 1)
combo['Name'] = combo.index.values
combo.index = range(len(combo))
away = combo[combo.Loc == 'Away'].index
temp = combo.loc[away,'Team']
combo.loc[away,'Team'] = combo.loc[away,'Opp']
combo.loc[away,'Opp'] = temp
combo.set_index('Name')
compare = combo if compare is None else compare.append(combo)
compare.loc[:,'date'] = pd.to_datetime(compare.date)
return compare
def boxStats(date):
try:
games = get_games(date)
url_base = 'https://www.basketball-reference.com/boxscores/{}0{}.html'
stats = None
for home, away in games:
t = pd.read_html(url_base.format(date.strftime('%Y%m%d'),home))
for idx in [0, int(len(t)/2)]:
temp = t[idx]
temp.columns = temp.columns.droplevel(0)
temp = temp.set_index('Starters').drop('Reserves').drop('Team Totals').fillna(0)
if idx == 0:
temp['Loc'] = 'Away'
temp['Team'] = away
temp['Opp'] = home
else:
temp['Loc'] = 'Home'
temp['Team'] = home
temp['Opp'] = away
stats = temp if stats is None else stats.append(temp)
stats.drop(stats[stats.MP.str[:3] == 'Did'].index, inplace=True)
stats.drop(stats[stats.MP.str[:3] == 'Not'].index, inplace=True)
stats['FP'] = stats.PTS.astype('int') + stats.TRB.astype('int') * 1.2 + stats.AST.astype('int') * 1.5 + stats.BLK.astype('int') * 3 + stats.STL.astype('int') * 3 - stats.TOV.astype('int')
stats.to_csv(format_fpath('stat', date))
except:
print('No games for {}'.format(date))
traceback.print_exc(file = sys.stdout)
def main():
url = "https://www.numberfire.com/nba/daily-fantasy/daily-basketball-projections"
soup = BeautifulSoup(requests.get(url).text, features="lxml")
table = soup.find_all("table")
row_marker = 0
data = []
for row in table[3].find_all("tr"):
columns = row.find_all("td")
data.append([x.get_text().strip() for x in columns])
cols = "Player,FP,Cost,Value,Min,Pts,Reb,Ast,Stl,Blk,TO".split(",")
df = pd.DataFrame(data, columns=cols).dropna()
player_data = df.Player.apply(parse_player)
df = df.join(
pd.DataFrame(
player_data.to_list(),
index=player_data.index,
columns=["Name", "Pos", "Team", "Opp", "Game"],
)).drop("Player", axis=1)
for pos in df.Pos.unique():
col = 'is{}'.format(pos)
df[col] = (df.Pos == pos).astype('int').to_numpy()
df.loc[:, 'Cost'] = df.Cost.apply(lambda x: int(''.join(x[1:].split(','))))
df.to_csv(format_fpath('proj'), index=False)
def generate(date=dt.date.today(), lineups=25, to_file=True):
data = get_proj(date)
fp_col = 'FP'
df = data.copy()
pos_mat = np.transpose(df.loc[:, ('isPG', 'isSG', 'isSF', 'isPF',
'isC')].to_numpy())
cur_proj = df[fp_col].copy()
b = np.array([2, 2, 2, 2, 1])
sal_max = 60000
x = cp.Variable(len(df), boolean=True)
salary_columns = 'Cost'
sal = df[salary_columns].to_numpy()
selections = None
for round in range(1, lineups + 1):
c = cur_proj.to_numpy()
objective = cp.Maximize(x.T @ c)
constraints = [pos_mat @ x == b, x >= 0, x <= 1,
x.T @ sal <= sal_max] #pos_mat @ x >= b_low
prob = cp.Problem(objective, constraints)
prob.solve(solver='GLPK_MI')
picks = df.iloc[x.value == 1].copy()
picks['round'] = round
selections = picks if selections is None else selections.append(picks)
cur_proj.loc[picks.index] = cur_proj.loc[picks.index].values * .95
if to_file:
selections.to_csv(format_fpath('line', date), index=False)
else:
return selections
def statRange(start_date, end_date):
stats = None
for date in pd.date_range(start_date,end_date):
try:
temp = pd.read_csv(format_fpath('stat',date.date()))
temp['Date'] = date
except:
continue
stats = temp if stats is None else stats.append(temp)
return stats
def build_feature_set(date = dt.date.today()):
proj = pd.read_csv(format_fpath('proj', date))
team_translation(proj)
teams = proj.Team.drop_duplicates()
hist = game_data(date - dt.timedelta(days = 1), 15)
offense, defense = calc_ratings(hist)
def_dict = defense.to_dict()
off_dict = offense.to_dict()
off_def = teams.apply(lambda x: off_dict[x]).rename('ortg').to_frame().join(teams.apply(lambda x: def_dict[x]).rename('drtg')).mean()
lineups = pd.read_csv(format_fpath('line', date))
team_translation(lineups)
lineups['ortg'] = off_def.ortg
lineups['drtg'] = off_def.drtg
lineups['Games'] = len(teams)/2
enc = oneHotTeams(defense)
defense_mat = enc.transform(np.reshape(lineups.Opp.to_numpy(),(-1,1)))
offense_mat = enc.transform(np.reshape(lineups.Team.to_numpy(),(-1,1)))
lineups['l_drtg'] = np.reshape(np.matmul(defense_mat,np.reshape(defense.values,(-1,1))),(-1))
lineups['l_ortg'] = np.reshape(np.matmul(offense_mat,np.reshape(offense.values,(-1,1))),(-1))
lineups['dscore'] = np.reshape(np.matmul(defense_mat, np.reshape(pd.read_csv(format_fpath('score',date - dt.timedelta(days = 1))).set_index('Defense').values,(-1,1))),(-1))
return lineups
def fp_score(cur_date, lookback):
stats = statRange(cur_date - dt.timedelta(days = lookback),cur_date)
df = game_data(cur_date,lookback)
df = df.join(df.index.to_frame()).set_index(['Date', 'Offense']).join(stats.groupby(['Date','Team']).sum().FP, on = ['Date', 'Offense']).dropna()
offense = df.groupby('Offense').mean().FP.rename('ortg')
defense = df.groupby('Defense').mean().FP.rename('drtg')
for _ in range(20):
mapper = df.join(offense, on = 'Offense').join(defense, on = 'Defense')
mapper['new-ortg'] = mapper.FP * 2 - mapper['drtg']
mapper['new-drtg'] = mapper.FP * 2 - mapper['ortg']
mapper.drtg = (mapper['new-drtg'] + mapper.drtg)/2
mapper.ortg = (mapper['new-ortg'] + mapper.ortg)/2
offense = mapper.groupby('Offense').mean().ortg
defense = mapper.groupby('Defense').mean().drtg
defense.to_csv(format_fpath('score',cur_date))
def get_proj(date):
df = pd.read_csv(format_fpath('proj', date))
df.loc[:, 'Name'] = df.Name.apply(lambda x: format_name(x))
mapper = player_team_map()
listed = df.Name.apply(lambda x: x in mapper.index)
df['PTeam'] = 'UNK'
df.loc[listed,
'PTeam'] = df.loc[listed].Name.apply(lambda x: mapper.loc[x]).values
df['Loc'] = df.apply(lambda x: 'Home' if x.PTeam == x.Team else 'Away',
axis=1)
df['Name'] = df.index.values
df.index = range(len(df))
away = df[df.Loc == 'Away'].index
temp = df.loc[away, 'Team']
df.loc[away, 'Team'] = df.loc[away, 'Opp']
df.loc[away, 'Opp'] = temp
df.set_index('Name')
return df