def get_fitted_gaussian_processor(X_train,
y_train,
constraint_upper,
standardize_y=True,
**gp_params):
# Initialize gaussian process regressor
gp = GaussianProcessRegressor()
gp.set_params(**gp_params)
logger.debug('Instantiated gaussian processor for objective function:\n' +
f'{gp}')
logger.debug(f"Fitting gaussian processor")
if standardize_y:
if constraint_upper is not None:
y_train = scale(np.hstack((y_train, constraint_upper)))
scaled_constraint_upper = y_train[-1]
y_train = y_train[:-1]
else:
y_train = scale([i for i in y_train])
scaled_constraint_upper = None
gp.constraint_upper = scaled_constraint_upper
else:
gp.constraint_upper = constraint_upper
logger.debug(f'X_train:\n{X_train}')
logger.debug(f'y_train\n{y_train}')
logger.debug(f'constraint_upper: {gp.constraint_upper}')
if gp_params is None or gp_params.get('alpha') is None:
# Find unique rows of X to avoid GP from breaking
ur = unique_rows(X_train)
gp.fit(X_train[ur], y_train[ur])
else:
gp.fit(X_train, y_train)
return gp
