Linear

class gpjax.kernels.location_functions.Linear(active_dims, weights=None, intercept=False)

Bases: AbstractLocationFunction

A location function that is log-linear in selected input columns.

\[\log g(x) = \beta^{\top} x_{\mathcal{A}} + b\]

Here \(x_{\mathcal{A}}\) are the columns in active_dims. The weights \(\beta\) start at zero, so a model starts as its stationary base kernel and moves away from it only as far as the data supports.

By default the function has no intercept (\(b = 0\)): the base kernel holds the overall scale, and the location function holds only the change over location. An intercept would duplicate the variance or lengthscale of the base kernel, and the data could not separate the two.

Standardise the covariate columns before fitting, so that the weights have similar scales.

Parameters:

• active_dims (list[int] | slice)

• weights (Any)

• intercept (bool)