import sqlite3
from sqlalchemy import create_engine
import time
time_sleep = 2
#list of the competition available on throme.ai (10/12/2017)
#competitions = ['English Championship','English Premier League','Italian Serie A','Spanish La Liga','NBA','NFL','NHL']
competitions = ['Italian Serie A']
#ThroneUser = "******"
#ThroneToken = "bcebc6d7-f224-4940-a816-74e3e6d4c34a" # Julien
ThroneUser = "******"
ThroneToken = "802f4123-4696-4b35-9e31-e8c32d91f658" #Nico
# API throne.ai :
throne = peyton.Throne(username=ThroneUser, token=ThroneToken)
def ThroneData(df):
df.reset_index()
#df.drop('Unnamed: 0', 1, inplace=True)
df['year'] = df.date.str.slice(0, 4)
df['month'] = df.date.str.slice(5, 7)
df['day'] = df.date.str.slice(8, 10)
# Get data to predict for each competitions. Is it possible to loop with the throttling limits?
#columns = Throne_historical_data.columns.tolist()
#print(columns)
i = 0
for compet in competitions:
def GetFrameWithFeatures():
# get competition data via peyton
throne = peyton.Throne(username='******',
token=os.environ['THRONE_TOKEN'])
# Get historical data for a competition
throne.competition('English Premier League').get_historical_data()
df = throne.competition.historical_data
# Get competition data for a competition
throne.competition('English Premier League').get_competition_data()
df_comp = throne.competition.competition_data
#df = pd.read_csv('english_premier_league_historical_data.csv')
#df_comp = pd.read_csv('english_premier_league_competition_data.csv')
# push additional columns
df['team_1_prob'] = np.nan
df['team_2_prob'] = np.nan
df['team_tie_prob'] = np.nan
df['data_type'] = 'hist'
df_comp['team_1_score'] = np.nan
df_comp['team_2_score'] = np.nan
df_comp['data_type'] = 'comp'
# bring together dfs
df = pd.concat([df, df_comp], axis=0)
# sort by time.
df = df.sort_values(['date'])
# get team list
teams = list(
set(df['team_1_name'].unique().tolist())
| set(df['team_2_name'].unique().tolist()))
#tmp_team_frame
team_frames = []
# break out dataframe by team (manual group by for ability to do complex feature construction)
for team in tqdm(teams):
# build tmp frame for all times team_1 appears
tmp_team_1 = df[df['team_1_name'] == team].copy()
tmp_team_1 = tmp_team_1[[
'id', 'date', 'team_1_score', 'team_2_score', 'team_2_name'
]]
tmp_team_1.columns = [
'id', 'date', 'team_score', 'opponent_score', 'opponent_name'
]
tmp_team_1['team_type'] = 'team_name_1'
# build tmp frame for all times team_1 appears
tmp_team_2 = df[df['team_2_name'] == team].copy()
tmp_team_2 = tmp_team_2[[
'id', 'date', 'team_2_score', 'team_1_score', 'team_1_name'
]]
tmp_team_2.columns = [
'id', 'date', 'team_score', 'opponent_score', 'opponent_name'
]
tmp_team_2['team_type'] = 'team_name_2'
# concatenate these frames and append
tmp_team_frame = pd.concat([tmp_team_1, tmp_team_2])
tmp_team_frame['team_name'] = team
if len(tmp_team_frame) > 20:
# build features
tmp_team_frame = build_rolling_features(tmp_team_frame)
# break out dataframe by team (manual group by for ability to do complex feature construction)
team_frames.append(tmp_team_frame)
df['result'] = df[['team_1_score', 'team_2_score']].apply(
lambda x: win_tie_loss(x['team_1_score'], x['team_2_score']), 1)
team_frame = pd.concat(team_frames, axis=0)
team_frame['result'] = team_frame[['team_score', 'opponent_score']].apply(
lambda x: win_tie_loss(x['team_score'], x['opponent_score']), 1)
# bring all team features together
#team_frame[['id','match_count','opponent_last_score','opponent_score_ema_10','opponent_score_ma_10','team_last_score','team_score_ema_10','team_score_ma_10','team_win_index','team_loss_index','tie_index']]
team_frame_1 = team_frame[[
'id', 'team_name', 'match_count', 'opponent_last_score',
'opponent_score_ema_10', 'opponent_score_ma_10', 'team_last_score',
'team_score_ema_10', 'team_score_ma_10', 'team_win_index',
'team_loss_index', 'tie_index'
]].copy()
team_frame_1.columns = [
'team_1_' + x if index > 1 else x
for index, x in enumerate(team_frame_1.columns.tolist())
]
team_frame_2 = team_frame[[
'id', 'team_name', 'match_count', 'opponent_last_score',
'opponent_score_ema_10', 'opponent_score_ma_10', 'team_last_score',
'team_score_ema_10', 'team_score_ma_10', 'team_win_index',
'team_loss_index', 'tie_index'
]].copy()
team_frame_2.columns = [
'team_2_' + x if index > 1 else x
for index, x in enumerate(team_frame_2.columns.tolist())
]
#team_frame
#len(df)
df = pd.merge(df,
team_frame_1,
left_on=['id', 'team_1_name'],
right_on=['id', 'team_name'])
df = pd.merge(df,
team_frame_2,
left_on=['id', 'team_2_name'],
right_on=['id', 'team_name'])
df['x_year'] = df['date'].apply(lambda x: year_fourier_components(x)[0])
df['y_year'] = df['date'].apply(lambda x: year_fourier_components(x)[1])
df['x_week'] = df['date'].apply(lambda x: week_fourier_components(x)[0])
df['y_week'] = df['date'].apply(lambda x: week_fourier_components(x)[1])
df['x_day'] = df['date'].apply(lambda x: day_fourier_components(x)[0])
df['y_day'] = df['date'].apply(lambda x: day_fourier_components(x)[1])
df
df[[
'id', 'data_type', 'result', 'team_1_score', 'team_2_score',
'is_february', 'is_november', 'c_ability_3', 'd_ability_1',
'd_ability_3', 'd_ability_4', 'd_form_4', 'd_h2h_2', 'team_1_name',
'team_2_name', 'team_1_match_count', 'team_1_opponent_last_score',
'team_1_opponent_score_ema_10', 'team_1_opponent_score_ma_10',
'team_1_team_last_score', 'team_1_team_score_ema_10',
'team_1_team_score_ma_10', 'team_1_team_win_index',
'team_1_team_loss_index', 'team_1_tie_index', 'team_2_match_count',
'team_2_opponent_last_score', 'team_2_opponent_score_ema_10',
'team_2_opponent_score_ma_10', 'team_2_team_last_score',
'team_2_team_score_ema_10', 'team_2_team_score_ma_10',
'team_2_team_win_index', 'team_2_team_loss_index', 'team_2_tie_index',
'x_year', 'y_year', 'x_week', 'y_week', 'x_day', 'y_day'
]].to_csv('epl_data_w_features.csv', index=False)
return (df)
'team_1_score_pred_lin', 'team_2_score_pred_lin',
'team_1_score_pred_tree', 'team_2_score_pred_tree'
], comp_data_ens)
comp_data["confidence"] = 1.0
## subbmit comp data
# check competition results
submit_frame = comp_data[[
'id', 'team_1_name', 'team_2_name', 'team_1_prob', 'confidence',
'team_tie_prob', 'team_2_prob'
]]
submit_frame
# manual confidence lever settings.
#submit_frame.set_value(5219,'confidence', 0.5)
#submit_frame.set_value(5221,'confidence', 0.25)
#submit
throne = peyton.Throne(username='******',
token=os.environ['THRONE_TOKEN'])
throne.competition('English Premier League').submit(submit_frame)
obj = {'hist_data': hist_data, 'features_list': feature_list}
import pickle
with open('tmp.pkl', 'wb') as f:
pickle.dump(obj, f)
import peyton
throne = peyton.Throne(username='******',
token="bcebc6d7-f224-4940-a816-74e3e6d4c34a")
# Get historical data for a competition
throne.competition('Italian Serie A').get_historical_data()
my_historical_data = throne.competition.historical_data
# Get competition data for a competition
throne.competition('Italian Serie A').get_competition_data()
my_competition_data = throne.competition.competition_data
print(my_competition_data)