Hi all,

Contrary to what I expected, BayesianRidge is not Bayesian w.r.t. the intercept parameter. This leads to the undesired outcome that all lines sampled from the parameter posterior go through the same point at the input x=0.

A manual workaround is to add a 1 column to the input and set fit_intercept=False. Then coef_ has the shape (d+1,) and sigma_ the shape (d+1, d+1).

Maybe this can be integrated into the current API. If there's an API change that's sensible, I'm happy to file a PR.

Thanks