def compute_gpy_model_rmse(model: GP, x_test: np.ndarray,
y_test: np.ndarray) -> float:
"""RMSE of a GPy model.
:param model:
:param x_test:
:param y_test:
:return:
"""
mean, var = model.predict(x_test)
y_pred = mean
return np.sqrt(mean_squared_error(y_test, y_pred))
def __init__(self, X: np.ndarray, Y: np.ndarray, kernel: GPy.kern.Kern, n_fidelities: int, mf_type: str,
likelihood: GPy.likelihoods.Likelihood=None):
"""
:param X: Training data features with fidelity input appended as last column
:param Y: Training data targets
:param kernel: Multi-fidelity kernel
:param n_fidelities: Number of fidelities in problem
:param likelihood: GPy likelihood object.
Defaults to MixedNoise which has different noise levels for each fidelity
"""
# Input checks
if not isinstance(X, np.ndarray):
raise ValueError('X should be an array')
if not isinstance(Y, np.ndarray):
raise ValueError('Y should be an array')
if X.ndim != 2:
raise ValueError('X should be 2d')
if Y.ndim != 2:
raise ValueError('Y should be 2d')
if np.any(X[:, -1] >= n_fidelities):
raise ValueError('One or more points has a higher fidelity index than number of fidelities')
# Make default likelihood as different noise for each fidelity
if likelihood is None:
likelihood = GPy.likelihoods.mixed_noise.MixedNoise(
[GPy.likelihoods.Gaussian(variance=1.) for _ in range(n_fidelities)])
y_metadata = {'output_index': X[:, -1].astype(int)}
if mf_type == 'autoregressive':
GP(X, Y, kernel, likelihood)
elif mf_type == 'recursive':
multiGP(X=X, Y=Y, kernel=kernel, Y_metadata=y_metadata)
else:
raise ValueError('mf_type must be autoregressive or recursive')
def log_likelihood(self):
return GP.log_likelihood(self)
def _log_likelihood_gradients(self):
dL_df = self.kern.dK_dX(self.dL_dK, self.X)
dL_dtheta = self.mapping.df_dtheta(dL_df, self.likelihood.Y)
return np.hstack(
(dL_dtheta.flatten(), GP._log_likelihood_gradients(self)))
def _set_params(self, x):
self.mapping._set_params(x[:self.mapping.num_params])
self.X = self.mapping.f(self.likelihood.Y)
GP._set_params(self, x[self.mapping.num_params:])
def _get_params(self):
return np.hstack((self.mapping._get_params(), GP._get_params(self)))
def _get_param_names(self):
return self.mapping._get_param_names() + GP._get_param_names(self)