Skip to main content

BayesianRidge

Bayesian ridge regression: a Gaussian linear model whose noise and weight precisions are estimated from the data by evidence maximisation, yielding predictive uncertainty for free.

Algorithm​

Gamma hyperpriors on the noise precision (alpha) and weight precision (lambda) are optimised by the iterative evidence-maximisation scheme of Tipping (2001). predict(X, with_std) returns the predictive mean and standard deviation.

Constructor​

Skigen::BayesianRidge<Scalar> model(int max_iter = 300, Scalar tol = 1e-3, Scalar alpha_1 = 1e-6, ...);

Parameters​

ParameterDefaultDescription
max_iter300Evidence-maximisation iterations.
tol1e-3Convergence tolerance.
alpha_1 / alpha_21e-6Gamma prior on the noise precision.
lambda_1 / lambda_21e-6Gamma prior on the weight precision.
compute_scorefalseTrack the log marginal likelihood.

Methods​

MethodDescription
fit(X, y)Optimise the evidence.
predict(X)Predictive mean.
predict(X, with_std)Mean and predictive std-dev.

Fitted Attributes​

AccessorDescription
coef()Posterior mean weights.
alpha()Estimated noise precision.
lambda_()Estimated weight precision.

Example​

Skigen::BayesianRidge<double> br;
br.fit(X, y);
auto [mean, std] = br.predict(X_test, Skigen::with_std);
Verified against scikit-learn

This estimator is checked by the parity suite. See the generator tests/parity/generate_bayesian_linear_reference.py and the reference fixtures in tests/parity/data/bayesian_ridge/, exercised by tests/parity/parity_bayesian_linear.cpp.

API Reference

For full signatures see the BayesianRidge API Reference.