gplvm_aep_examples.py 文件源码

def run_mnist():
    np.random.seed(42)

    # import dataset
    f = gzip.open('./tmp/data/mnist.pkl.gz', 'rb')
    (x_train, t_train), (x_valid, t_valid), (x_test, t_test) = cPickle.load(f)
    f.close()

    Y = x_train[:100, :]
    labels = t_train[:100]

    Y[Y < 0.5] = -1
    Y[Y > 0.5] = 1

    # inference
    print "inference ..."
    M = 30
    D = 2
    # lvm = aep.SGPLVM(Y, D, M, lik='Gaussian')
    lvm = aep.SGPLVM(Y, D, M, lik='Probit')
    # lvm.train(alpha=0.5, no_epochs=10, n_per_mb=100, lrate=0.1, fixed_params=['sn'])
    lvm.optimise(method='L-BFGS-B', alpha=0.1)
    plt.figure()
    mx, vx = lvm.get_posterior_x()
    zu = lvm.sgp_layer.zu
    plt.scatter(mx[:, 0], mx[:, 1], c=labels)
    plt.plot(zu[:, 0], zu[:, 1], 'ko')

    nx = ny = 30
    x_values = np.linspace(-5, 5, nx)
    y_values = np.linspace(-5, 5, ny)
    sx = 28
    sy = 28
    canvas = np.empty((sx * ny, sy * nx))
    for i, yi in enumerate(x_values):
        for j, xi in enumerate(y_values):
            z_mu = np.array([[xi, yi]])
            x_mean, x_var = lvm.predict_f(z_mu)
            t = x_mean / np.sqrt(1 + x_var)
            Z = 0.5 * (1 + special.erf(t / np.sqrt(2)))
            canvas[(nx - i - 1) * sx:(nx - i) * sx, j *
                   sy:(j + 1) * sy] = Z.reshape(sx, sy)

    plt.figure(figsize=(8, 10))
    Xi, Yi = np.meshgrid(x_values, y_values)
    plt.imshow(canvas, origin="upper", cmap="gray")
    plt.tight_layout()

    plt.show()