type="test",
fold=str(i),
verbose=False)[1].values())
manager = ModelManager()
manager.assign_sets(train=train)
tup = manager.create_mask(
train.iloc[:, :-1],
global_dirs.variable_selection[0],
select=global_dirs.variable_selection[1]
) # This tuple shouldn't take care about y_column index
scalers = manager.preprocess_train(tup, scale_Y=False)
dnn_model = manager.fit_dnn_regression([(9, 'relu'), (18, 'relu'),
(56, 'relu'), (11, 'relu'),
(10, 'relu')],
epochs=300,
batch_size=30,
use_dropout=False)
results = manager.predict_dnn_regression(test, tup)
#X_train = train.loc[:, train.columns != train.columns[-1]]
#X_test = test.loc[:, test.columns != test.columns[-1]]
#var_selection = reader.create_mask(X_train, global_dirs.variable_selection[0], select=global_dirs.variable_selection[1])
#X_train = X_train.loc[:, var_selection]
#X_test = X_test.loc[:, var_selection]
#y_train = train.loc[:, train.columns[-1]]
#y_test = test.loc[:, test.columns[-1]]
if not os.path.isdir(global_dirs.results_path):
os.mkdir(global_dirs.results_path)
if not os.path.isdir(global_dirs.dnn_path):
os.mkdir(global_dirs.dnn_path)
if not os.path.isdir(global_dirs.dnn_path + "scalers/"):
os.mkdir(global_dirs.dnn_path + "scalers/")
if not os.path.isdir(global_dirs.dnn_path + "model/"):
os.mkdir(global_dirs.dnn_path + "model/")
if not os.path.isdir(global_dirs.dnn_path + "results/"):
os.mkdir(global_dirs.dnn_path + "results/")
dnn_model = manager.fit_dnn_regression(
[(9, 'relu'), (18, 'relu'), (56, 'relu'), (11, 'relu'), (10, 'relu')],
epochs=500,
batch_size=30,
save_dir=global_dirs.dnn_path + "model/",
save=True,
use_dropout=False)
if (isinstance(scalers, tuple)):
joblib.dump(scalers[0], global_dirs.dnn_path + "scalers/scaler_X.h5")
joblib.dump(scalers[1], global_dirs.dnn_path + "scalers/scaler_y.h5")
else:
joblib.dump(scalers, global_dirs.dnn_path + "scalers/scaler_X.h5")
test = {
name.split("-")[1]: data
for name, data in manager.read_data(global_dirs.splitted_data_path,
formats=["hdf"],
sim="qgsjet",
