Function plot_regression

numpy_ml/plots/lm_plots.py:219–281

()

def plot_regression():
    np.random.seed(12345)

    fig, axes = plt.subplots(4, 4)
    for i, ax in enumerate(axes.flatten()):
        n_in = 1
        n_out = 1
        n_ex = 50
        std = np.random.randint(0, 100)
        intercept = np.random.rand() * np.random.randint(-300, 300)
        X_train, y_train, X_test, y_test, coefs = random_regression_problem(
            n_ex, n_in, n_out, intercept=intercept, std=std, seed=i
        )

        LR = LinearRegression(fit_intercept=True)
        LR.fit(X_train, y_train)
        y_pred = LR.predict(X_test)
        loss = np.mean((y_test - y_pred) ** 2)
        r2 = r2_score(y_test, y_pred)

        LR_var = BayesianLinearRegressionKnownVariance(
            mu=np.c_[intercept, coefs][0],
            sigma=np.sqrt(std),
            V=None,
            fit_intercept=True,
        )
        LR_var.fit(X_train, y_train)
        y_pred_var = LR_var.predict(X_test)
        loss_var = np.mean((y_test - y_pred_var) ** 2)
        r2_var = r2_score(y_test, y_pred_var)

        LR_novar = BayesianLinearRegressionUnknownVariance(
            alpha=1, beta=2, mu=np.c_[intercept, coefs][0], V=None, fit_intercept=True,
        )
        LR_novar.fit(X_train, y_train)
        y_pred_novar = LR_novar.predict(X_test)
        loss_novar = np.mean((y_test - y_pred_novar) ** 2)
        r2_novar = r2_score(y_test, y_pred_novar)

        xmin = min(X_test) - 0.1 * (max(X_test) - min(X_test))
        xmax = max(X_test) + 0.1 * (max(X_test) - min(X_test))
        X_plot = np.linspace(xmin, xmax, 100)
        y_plot = LR.predict(X_plot)
        y_plot_var = LR_var.predict(X_plot)
        y_plot_novar = LR_novar.predict(X_plot)

        ax.scatter(X_test, y_test, marker="x", alpha=0.5)
        ax.plot(X_plot, y_plot, label="linear regression", alpha=0.5)
        ax.plot(X_plot, y_plot_var, label="Bayes (w var)", alpha=0.5)
        ax.plot(X_plot, y_plot_novar, label="Bayes (no var)", alpha=0.5)
        ax.legend()
        ax.set_title(
            "MSE\nLR: {:.2f} Bayes (w var): {:.2f}\nBayes (no var): {:.2f}".format(
                loss, loss_var, loss_novar
            )
        )

        ax.xaxis.set_ticklabels([])