abstract_process.py 文件源码

def predict(self, X_pred, covariance=False):
        """Leverage Bayesian posterior inference to compute the predicted mean
        and variance of a given set of inputs given the available training data.
        Notice that it is necessary to first fit the Gaussian process model
        before posterior inference can be performed.
        """
        # Compute the cross covariance between training and the requested
        # inference locations. Also compute the covariance matrix of the
        # observed inputs and the covariance at the inference locations.
        if type(self.kernel) == SumKernel:
            K_pred = self.kernel.cov(X_pred, include_noise=False)
        else:
            K_pred = self.kernel.cov(X_pred)
        K_cross = self.kernel.cov(X_pred, self.X)
        v = spla.solve_triangular(self.L, K_cross.T, lower=True)
        # Posterior inference. Notice that we add a small amount of noise to the
        # diagonal for regulatization purposes.
        mean = K_cross.dot(self.alpha)
        cov = self.predict_prefactor * (
            K_pred - v.T.dot(v) + 1e-8 * np.eye(K_pred.shape[0])
        )
        # Compute the diagonal of the covariance matrix if we wish to disregard
        # all of the covariance information and only focus on the variances at
        # the given inputs.
        if covariance:
            return mean, cov
        else:
            return mean, np.sqrt(np.diag(cov))