Function plot_gp_dist

numpy_ml/plots/nonparametric_plots.py:196–238

()

def plot_gp_dist():
    np.random.seed(12345)
    sns.set_context("paper", font_scale=0.95)

    X_test = np.linspace(-10, 10, 100)
    X_train = np.array([-3, 0, 7, 1, -9])
    y_train = np.sin(X_train)

    fig, axes = plt.subplots(1, 3)
    G = GPRegression(kernel="RBFKernel", alpha=0)
    G.fit(X_train, y_train)

    y_pred_prior = G.sample(X_test, 3, "prior")
    y_pred_posterior = G.sample(X_test, 3, "posterior_predictive")

    for prior_sample in y_pred_prior:
        axes[0].plot(X_test, prior_sample.ravel(), lw=1)
    axes[0].set_title("Prior samples")
    axes[0].set_xticks([])
    axes[0].set_yticks([])

    for post_sample in y_pred_posterior:
        axes[1].plot(X_test, post_sample.ravel(), lw=1)
    axes[1].plot(X_train, y_train, "ko", ms=1.2)
    axes[1].set_title("Posterior samples")
    axes[1].set_xticks([])
    axes[1].set_yticks([])

    y_pred, conf = G.predict(X_test)

    axes[2].plot(X_test, np.sin(X_test), lw=1, label="true function")
    axes[2].plot(X_test, y_pred, lw=1, label="MAP estimate")
    axes[2].fill_between(X_test, y_pred + conf, y_pred - conf, alpha=0.1)
    axes[2].plot(X_train, y_train, "ko", ms=1.2, label="observed")
    axes[2].legend(fontsize="x-small")
    axes[2].set_title("Posterior mean")
    axes[2].set_xticks([])
    axes[2].set_yticks([])

    fig.set_size_inches(6, 2)
    plt.tight_layout()
    plt.savefig("img/gp_dist.png", dpi=300)
    plt.close("all")