('data/original/wc_2010_games_real.csv',
'data/original/wc_2010_bets.csv', "2010-06-11")]
feature_sets = [("all_features", all_features),
("general_features", other_features),
("player_features", player_features)]
file_name = "outcome_report_full.txt"
reports = []
for (name, feature_set) in feature_sets:
write_log(file_name, str(datetime.datetime.now()))
write_log(file_name,
f"Running test for feature set: {name}",
print_text=True)
data_loader = DataLoader(feature_set)
X, y = data_loader.get_all_data("home_win")
arguments = get_grid_search_arguments(X)
results = run_grid_search(arguments, X, y)
results.to_csv(f"gboost_hyperparam_optimization_{name}.csv")
best_params_dict = get_best_params(results)
optimal_params = {'n_estimators': 250}
optimal_params["learning_rate"] = best_params_dict["learning_rate"]
optimal_params["max_depth"] = best_params_dict["max_depth"]
optimal_params["min_samples_leaf"] = best_params_dict["min_samples_leaf"]
optimal_params["max_features"] = best_params_dict["max_features"]
write_log(file_name, str(optimal_params), print_text=True)
for (tt_file, bet_file, filter_start) in tournament_parameters:
]
feature_sets = [
# ("all_features", all_features, "score_hyperparam_optimization_all_features.csv"),
# ("general_features", other_features, "score_hyperparam_optimization_general_features.csv"),
# ("player_features", player_features, "score_hyperparam_optimization_player_features.csv")
("rfe_features", rfe_feature, "score_hyperparam_optimization_rfe.csv"))
]
file_name = "score_report_full.txt"
reports = []
for (name, feature_set, fname) in feature_sets:
write_log(file_name, str(datetime.datetime.now()))
write_log(file_name, f"Running test for feature set: {name}", print_text=True)
data_loader = DataLoader(feature_set)
params = get_default_parameters()
if os.path.isfile(fname):
write_log(file_name, f"Hyperparameters found for: {name}", print_text=True)
results = pd.read_csv(fname)
else:
Xhome, yhome, Xaway, yaway = data_loader.get_all_data(["home_score", "away_score"])
_, outcomes = data_loader.get_all_data("home_win")
arguments = get_cv_grid_search_arguments(params, Xhome)
results = run_grid_search_for_score(arguments, Xhome, yhome, Xaway, yaway, outcomes)
results.to_csv(f"score_hyperparam_optimization_{name}.csv")
best_params_dict = get_best_params(results)
write_log(file_name, str(best_params_dict), print_text=True)
params = {
'oob_score' : True,
'bootstrap': True,
'n_jobs':-1,
'n_estimators': 1000,
"max_features": "sqrt",
"max_depth": 8,
"min_samples_leaf": 3
}
avg_accuracies = []
avg_log_lossss = []
features = []
while len(feature_set) > 0:
data_loader = DataLoader(feature_set)
accuracies = []
log_losses = []
feature_values = {}
for i in range(100):
model = RandomForestClassifier(**params)
X_train, y_train, X_test, y_test = data_loader.get_train_and_test_dataset("home_win", random_state=None)
model.fit(X_train, y_train)
y_true, y_pred = y_test, model.predict(X_test)
accuracies.append(accuracy_score(y_true, y_pred))
y_true, y_prob = y_test, model.predict_proba(X_test)
log_losses.append(log_loss(y_true, y_prob))
if args.y == 2010:
tt_file = 'data/original/wc_2010_games_real.csv'
mb_file = 'data/original/wc_2010_bets.csv'
filter_start = "2010-06-11"
elif args.y == 2014:
tt_file = 'data/original/wc_2014_games_real.csv'
mb_file = 'data/original/wc_2014_bets.csv'
filter_start = "2014-06-12"
else:
tt_file = 'data/original/wc_2018_games_real.csv'
mb_file = 'data/original/wc_2018_bets.csv'
filter_start = "2018-06-13"
prefix = f"{args.f}_{args.y}"
dl = DataLoader(all_features, filter_start=filter_start)
model_parameters = get_default_parameters()
model_parameters["max_depth"] = 8
model_parameters["max_features"] = "sqrt"
model_parameters["min_samples_leaf"] = 1
af_data = simulate(tt_file, mb_file, dl, model_parameters,
f"{prefix}_all_features")
dl = DataLoader(other_features, filter_start=filter_start)
model_parameters["max_depth"] = 8
model_parameters["max_features"] = "log2"
model_parameters["min_samples_leaf"] = 10
gf_data = simulate(tt_file, mb_file, dl, model_parameters,
f"{prefix}_general_features")
dl = DataLoader(player_features, filter_start=filter_start)