def make_output():
tourney_seeds = read_csv_as_np_array('../Data/tourney_seeds.csv',
header=False)
neutralized_season_data = neutralize_season_data(
'../Data/regular_season_detailed_results.csv',
'../Data/tourney_detailed_results.csv',
'../Data/neutralized_season_data.csv'
)
tuning_params = dict(std=0.05)
output = [['id', 'pred']]
for season in ['2011', '2012', '2013', '2014']:
tourney_teams = np.sort(tourney_seeds[np.where(tourney_seeds[:, 0] == season)[0]][:, 2])
while len(tourney_teams) > 1:
team1 = tourney_teams[0]
tourney_teams= np.delete(tourney_teams, 0)
for team2 in tourney_teams:
prob = calc_new_features(
neutralized_season_data,
season,
team1,
team2,
tuning_params
)[0]
game_id = [season+'_'+team1+'_'+team2, str(prob)]
output.append(game_id)
print game_id
output_file = open('../Data/out.csv', 'wb')
csv.writer(output_file).writerows(output)
def make_output_2015():
tourney_seeds = read_csv_as_np_array('../Data/tourney_seeds_2015.csv',
header=False)
neutralized_season_data = neutralize_season_data(
'../Data/regular_season_detailed_results_combined.csv',
'../Data/tourney_detailed_results.csv',
'../Data/neutralized_season_data_combined.csv'
)
tuning_params = dict(std=0.05)
output = [['id', 'pred', 'score1', 'score2']]
for season in ['2015']:
tourney_teams = np.sort(tourney_seeds[np.where(tourney_seeds[:, 0] == season)[0]][:, 2])
while len(tourney_teams) > 1:
team1 = tourney_teams[0]
tourney_teams= np.delete(tourney_teams, 0)
for team2 in tourney_teams:
features = calc_new_features(
neutralized_season_data,
season,
team1,
team2,
tuning_params
)
prob = features[0]
score1 = np.round(features[2]).astype(int)
score2 = np.round(features[3]).astype(int)
game_id = [season+'_'+team1+'_'+team2, str(prob), str(score1), str(score2)]
output.append(game_id)
print game_id
output_file = open('../Data/out_2015_with_scores.csv', 'wb')
csv.writer(output_file).writerows(output)
def main_2015_03_12():
tourney_data = read_csv_as_np_array('../Data/tourney_detailed_results.csv',
header=False)
neutralized_season_data = neutralize_season_data(
'../Data/regular_season_detailed_results.csv',
'../Data/tourney_detailed_results.csv',
'../Data/neutralized_season_data.csv'
)
# test on 2011-2014 tourney results
test_data_min_index = np.min(np.where(tourney_data[:, 0] == '2014')[0])
train_data = tourney_data
for std in [0.05]: # np.linspace(0.01, 0.1, 10):
tuning_params = dict(std=std)
#scramble the results to eliminate fact that team1 is always winner in tourney data
results = np.random.randint(0, 2, len(train_data))
features = []
output = [['id', 'pred']]
for i in xrange(len(train_data)):
season = train_data[i, 0]
if results[i] == 1:
team1 = train_data[i, 2]
team2 = train_data[i, 4]
else:
team1 = train_data[i, 4]
team2 = train_data[i, 2]
features.append(
calc_new_features(
neutralized_season_data,
season,
team1,
team2,
tuning_params
)
)
if int(season) >= 2011:
if int(team1) < int(team2):
game_id = [season+'_'+team1+'_'+team2, str(features[-1][0])]
else:
game_id = [season+'_'+team2+'_'+team1, str(1.0-features[-1][0])]
output.append(game_id)
# output_file = open('../Data/out.csv', 'wb')
# csv.writer(output_file).writerows(output)
features = np.array(features)
for season in np.unique(train_data[:, 0]): # ['2011', '2012', '2013', '2014']:
idx = np.where(train_data[:, 0] == season)[0]
log_loss = -np.mean(results[idx]*np.log(features[idx, 0])
+ (1-results[idx])*np.log(1.0 - features[idx, 0]))
junk = 0.98*(features[idx, 0]-features[idx, 0].min())/(features[idx, 0].max() - features[idx, 0].min()) + 0.01
junk_log_loss = -np.mean(results[idx]*np.log(junk)
+ (1-results[idx])*np.log(1.0 - junk))
print std, season, log_loss, junk_log_loss
