loglik = new(loglik_gauss, om, terms, y, x)This is a standard model which has the form
$$y = \langle \phi(x), \theta \rangle + \varepsilon, \varepsilon \sim
N(0,\sigma^2)$$
where \(\phi(x)\) is the basis, \(\theta\) is the coefficient vector,
\(\varepsilon\) is an unseen noise vector.
The parameter vector is of length 1 where
para \(= \log(\sigma)\). It is a faster (sometimes) version of
loglik_std but can only handle diagonal variational
inference.
Arguments
- om
an
outermodinstance to be referred to- terms
a matrix of
terms, must have as many columns as dims inom- y
a vector of observations
- x
a matrix of predictors, must have as many columns as dims in
omand the same number of rows asy
See also
base class: lpdf