This is a base class designed to handle the learning of the underlying coefficients, hyperparameters, and parameters associated with a specific learning instance. Polymorphism allows for the implied methods to be used across several similar classes.
Fields
lpdf$valcurrent value
lpdf$paracurrent model parameters
lpdf$coeffcurrent coefficients
lpdf$compute_valon calling
update, compute value and store invallpdf$gradcurrent gradient with respect to coefficients
lpdf$gradhypcurrent gradient with respect to covariance hyperparameters
lpdf$gradparacurrent gradient with respect to model parameters
lpdf$compute_gradon calling
update, compute gradient with respect to coefficients and store ingradlpdf$compute_gradhypon calling
update, compute gradient with respect to covariance hyperparameters and store ingradhyplpdf$compute_gradparaon calling
update, compute gradient with respect to model parameters and store ingradparalpdf$update(coeff)update using new coefficients
lpdf$optcg(tol,epoch)do optimization with respect to coefficients via conjugate gradient
lpdf$optnewton()do optimization via matrix inversion, one Newton step
lpdf$updateom()update based on recent version of
outermodlpdf$updatepara(para)update using new model parameters
lpdf$updateterms(terms)update using new
termslpdf$hess()returns the hessian with respect to coefficients
lpdf$hessgradhyp()returns gradient of
hess()with respect to covariance hyperparameterslpdf$hessgradpara()returns the gradient of
hess()with respect to model parameterslpdf$diaghess()returns the diagonal of the hessian with respect to coefficients
lpdf$diaghessgradhyp()returns the gradient of
diaghess()with respect to covariance hyperparameterslpdf$diaghessgradpara()returns the gradient of
diaghess()with respect to model parameterslpdf$paralpdf(para)compute the log-prior on the parameters, useful for fitting
lpdf$paralpdf_grad(para)gradient of
paralpdf(para)
See also
container class: lpdfvec
derived classes: loglik_std,
loglik_gauss, loglik_gda,
logpr_gauss