gpr_ep_examples.py 文件源码

def run_regression_1D_pep_training(stoc=False):
    np.random.seed(42)

    print "create dataset ..."
    N = 200
    X = np.random.rand(N, 1)
    Y = np.sin(12 * X) + 0.5 * np.cos(25 * X) + np.random.randn(N, 1) * 0.2
    # plt.plot(X, Y, 'kx', mew=2)

    def plot(m):
        xx = np.linspace(-0.5, 1.5, 100)[:, None]
        mean, var = m.predict_f(xx)
        zu = m.sgp_layer.zu
        mean_u, var_u = m.predict_f(zu)
        plt.figure()
        plt.plot(X, Y, 'kx', mew=2)
        plt.plot(xx, mean, 'b', lw=2)
        plt.fill_between(
            xx[:, 0],
            mean[:, 0] - 2 * np.sqrt(var[:, 0]),
            mean[:, 0] + 2 * np.sqrt(var[:, 0]),
            color='blue', alpha=0.2)
        plt.errorbar(zu, mean_u, yerr=2 * np.sqrt(var_u), fmt='ro')
        plt.xlim(-0.1, 1.1)

    # inference
    print "create model and optimize ..."
    M = 20
    alpha = 0.1
    model_pep = pep.SGPR_rank_one(X, Y, M, lik='Gaussian')
    if stoc:
        mb_size = M
        fname = '/tmp/gpr_pep_reg_stoc.pdf'
        adam_lr = 0.005
    else:
        mb_size = N
        fname = '/tmp/gpr_pep_reg.pdf'
        adam_lr = 0.05
    model_pep.optimise(method='adam', mb_size=mb_size, adam_lr=adam_lr, alpha=alpha, maxiter=2000)
    plot(model_pep)
    plt.savefig(fname)