def evaluate(self, X, ytrue):
"""
Evaluate the model based on the RMSE
:type X: np.ndarray
:param X: The feature array.
:type ytrue: np.ndarray
:param ytrue: The true label array.
"""
ypred, dummy = self.predict(X)
return rmse(ytrue, ypred)
def evaluate(self, ytrue, mean_pred, std_pred=False):
"""
Negative - log -likelihood for the prediction of a gaussian probability
"""
if std_pred is None:
return rmse(ytrue, mean_pred)
else:
mean = mean_pred
sigma = std_pred + 1e-6 # adding 1-e6 for numerical stability reasons
first = 0.5 * np.log(np.square(sigma))
second = np.square(mean - ytrue) / (2 * np.square(sigma))
summed = first + second
loss = np.mean(summed, axis=-1)
return loss