tune_hyperparms_regression.py 文件源码

def compute_mar_likelihood(X_train, X_test, y_train, sigma, l):
    """
    compute log marginal likelihood for tuning parameters using Bayesian optimization
    :param X_train: training data
    :param X_test: test data
    :param y_train: training targets
    :param sigma: output variance
    :param l: lengthscalar
    :return: log marginal likelihood
    """
    s = 0.0005  # noise variance and zero mean for noise
    n = len(X_train)

    # choose RBF kernel in this regression case
    K_train = RBF_kernel(X_train, X_train, sigma, l)
    L = np.linalg.cholesky(K_train + s * np.eye(n))
    m = np.linalg.solve(L, y_train)
    alpha = np.linalg.solve(L.T, m)

    # compute log marginal likelihood
    log_marg_likelihood = -.5 * np.dot(y_train.T, alpha) - np.log(np.diagonal(L)).sum(0) - n / 2.0 * np.log(2 * np.pi)
    return log_marg_likelihood