Nothing
R/RcppExports.R
In outerbase: Outer Product Regression
# Generated by using Rcpp::compileAttributes() -> do not edit by hand
# Generator token: 10BE3573-1514-4C36-9D1C-5A225CD40393
#' @name setcovfs
#' @title Set covariance functions
#' @description
#' \preformatted{
#' setcovfs(om, covnames)
#' }
#' Sets the covariance functions for an outermod class instance.
#' This is first thing one does when creating an outermod instance.
#' @param om an \code{\link{outermod}} instance
#' @param covnames a vector of strings of the covariance functions
#' @examples
#' om = new(outermod)
#' setcovfs(om, c("mat25", "mat25", "mat25"))
#' setcovfs(om, c("mat25", "mat25pow", "mat25", "mat25ang"))
#' @seealso \code{\link{outermod}}
#' @returns no value is returned, \code{om} is updated
NULL
#' @name setknot
#' @title Set knot points
#' @description
#' \preformatted{
#' setknot(om, knotslist)
#' }
#' Sets the knot points of \code{om} to \code{knotslist} to estimate the
#' eigenfunctions and eigenvalues. It will naturally check if the knot points
#' have the same dimension as the covariance functions. It will also check if
#' the knot points are within reasonable bounds for the covariance functions.
#' @param om an \code{\link{outermod}} instance
#' @param knotslist a list of one dimensional vectors
#' @examples
#' om = new(outermod)
#' setcovfs(om, c("mat25", "mat25", "mat25"))
#' knotslist = list(seq(0,1,by=0.01),seq(0,1,by=0.01),seq(0,1,by=0.01))
#' setknot(om, knotslist)
#' @seealso \code{\link{outermod}}, \code{\link{setcovfs}}
#' @returns no value is returned, \code{om} is updated
NULL
#' @name gethyp
#' @title Get the hyperparameters
#' @description
#' \preformatted{
#' hyp = gethyp(om)
#' }
#' Gets the current hyperparameters from an \code{\link{outermod}} instance. It
#' formats them in a way that makes reading in \code{R} easier.
#' @param om an \code{\link{outermod}} instance
#' @examples
#' om = new(outermod)
#' setcovfs(om, c("mat25", "mat25", "mat25"))
#' hyp = gethyp(om)
#' print(hyp)
#' @returns a vector of parameters
#' @seealso \code{\link{outermod}}
NULL
#' @name getpara
#' @title Get the model parameters
#' @description
#' \preformatted{
#' para = getpara(logpdf)
#' }
#' This function gets the current parameters from an \code{\link{lpdf}} class
#' instance. It formats them in a way that makes reading in \code{R} easier.
#' @returns a vector of parameters
#' @param logpdf an \code{\link{lpdf}} class instance
NULL
#' @name outermod
#' @aliases
#' Rcpp_outermod-class Rcpp_outermod
#' @title Outer product-type model
#' @description This is a class used to construct \code{\link{outerbase}}
#' class instances. It stores key information for constructing a basis.
#' @field \link{outermod$updatehyp}(hyp) update hyperparameters
#' @field \link{outermod$selectterms}(numterms) find best \code{numterms} terms
#' @field \link{outermod$getvar}(terms) find variances of coefficients
#' associated with \code{terms}
#' @examples
#' om = new(outermod)
#' setcovfs(om, c("mat25", "mat25", "mat25"))
#' setknot(om,
#' list(seq(0,1,by=0.01),seq(0,1,by=0.01),seq(0,1,by=0.01)))
#' terms = om$selectterms(40)
#' coeffvar =om$getvar(terms)
#' hyp = gethyp(om)
#' hyp[1:2] = 0.5
#' om$updatehyp(hyp)
#' coeffvar = om$getvar(terms)
#' @seealso \code{\link{outerbase}} the main product from an outermod
#' @seealso \code{\link{setcovfs}}, \code{\link{setknot}},
#' \code{\link{gethyp}}
#' @returns no returns, this is a class which contains methods
NULL
#' @name outermod$updatehyp
#' @title Update hyperparameters
#' @description
#' \preformatted{
#' outermod$updatehyp(hyp)
#' }
#' Updates the hyperparameters for the instance of outermod.
#' @param hyp A vector of hyperparameters
#' @returns no value is returned, the class instance is updated
#' @seealso \code{\link{outermod}}
NULL
#' @name outermod$selectterms
#' @title Select optimal terms
#' @description
#' \preformatted{
#' terms = om$selectterms(numterms)
#' }
#' Returns the best \code{numterms} given \code{outermod} currently using
#' maximum variance criteria.
#' @param numterms number of basis \code{terms} desired
#' @returns a matrix of \code{terms}
#' @seealso \code{\link{outermod}}
NULL
#' @name outermod$getvar
#' @title Get variance of coefficients
#' @description
#' \preformatted{
#' coeffvar = outermod$getvar(terms)
#' }
#' Returns the variance of the coefficients associated with \code{terms}.
#' @param terms a matrix of \code{terms}
#' @returns a vector of variances of each coefficient
#' @seealso \code{\link{outermod}}
NULL
#' @name outerbase
#' @aliases
#' Rcpp_outerbase-class Rcpp_outerbase
#' @title Outer product-type basis
#' @description
#' \preformatted{
#' ob = new(outerbase, om, x)
#' }
#' Class that handles the basis for a given set of points
#' \code{x}.
#' @param x a matrix of predictors, must have as many columns as dims in
#' \code{om}
#' @field nthreads number of threads for \code{omp} to use
#' @field \link{outerbase$getbase}(k) to get each dimensions basis
#' functions
#' @field \link{outerbase$getmat}(terms) to get the basis matrix at
#' \code{terms}
#' @field \link{outerbase$build}() to (re)build the basis instance
#' @field \link{outerbase$matmul}(terms,a) matrix multiply without
#'building the basis matrix
#' @field \link{outerbase$tmatmul}(terms,a) transpose matrix multiply
#' without building the basis matrix
#' @examples
#' om = new(outermod)
#' setcovfs(om, c("mat25", "mat25", "mat25"))
#' setknot(om,
#' list(seq(0,1,by=0.025),seq(0,1,by=0.025),seq(0,1,by=0.025)))
#' x = matrix(runif(10*3),ncol=3)
#' ob = new(outerbase, om, x)
#' terms = om$selectterms(40)
#' basismat = ob$getmat(terms)
#' @seealso \code{\link{outermod}} the core element that controls outerbase
#' @returns no returns, this is a class which contains methods
NULL
#' @name outerbase$getbase
#' @title Get base functions
#' @description
#' \preformatted{
#' basis_func = outerbase$getbase(k)
#' }
#' Returns the basis for dimension \code{k}. Designed mostly for
#' visualization.
#' @param k An integer from that corresponds to the dimension.
#' @returns a matrix of evaluated basis functions
#' @seealso \code{\link{outerbase}}
NULL
#' @name outerbase$getmat
#' @title Get basis matrix
#' @description
#' \preformatted{
#' basismat = outerbase$getmat(terms)
#' }
#' Returns the basis matrix for a given set of \code{terms}.
#' @param terms a matrix of terms
#' @returns a matrix of evaluated basis functions based on
#' \code{terms}.
#' @seealso \code{\link{outerbase}}
NULL
#' @name outerbase$build
#' @title Builds the outerbase
#' @description
#' \preformatted{
#' outerbase$build()
#' }
#' Build (or re-build) a basis based on the recent evaluation
#' of \code{\link{outermod}}.
#' @seealso \code{\link{outerbase}}
#' @returns nothing is returned, the class instance is updated
NULL
#' @name outerbase$matmul
#' @title Matrix multiply
#' @description
#' \preformatted{
#' b = outerbase$matmul(terms, a)
#' }
#' Multiplies the basis times a vector without building the basis
#' matrix.
#' @param terms a matrix of \code{terms}
#' @param a a vector of length the same as the rows in \code{terms}
#' @returns a vector resulting from the matrix multiplication
#' @seealso \code{\link{outerbase}}
NULL
#' @name outerbase$tmatmul
#' @title Transpose Matrix multiply
#' @description
#' \preformatted{
#' b = outerbase$tmatmul(terms, a)
#' }
#' Multiplies the transpose of the basis times a vector without
#' building the basis matrix.
#' @param terms a matrix of \code{terms}
#' @param a a vector of length the same as the rows in \code{outerbase}
#' @returns a vector resulting from the matrix multiplication
#' @seealso \code{\link{outerbase}}
NULL
#' @name lpdf
#' @aliases
#' Rcpp_lpdf-class Rcpp_lpdf
#' @title Log probability density function class
#' @description 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.
#'
#' @field lpdf$val current value
#' @field lpdf$para current model parameters
#' @field lpdf$coeff current coefficients
#' @field lpdf$compute_val on calling \code{update}, compute value and store in
#' \code{val}
#' @field lpdf$grad current gradient with respect to coefficients
#' @field lpdf$gradhyp current gradient with respect to covariance hyperparameters
#' @field lpdf$gradpara current gradient with respect to model parameters
#' @field lpdf$compute_grad on calling \code{update}, compute gradient with
#' respect to coefficients and store in \code{grad}
#' @field lpdf$compute_gradhyp on calling \code{update}, compute gradient
#' with respect to covariance hyperparameters and store in \code{gradhyp}
#' @field lpdf$compute_gradpara on calling \code{update}, compute gradient
#' with respect to model parameters and store in \code{gradpara}
#' @field lpdf$update(coeff) update using new coefficients
#' @field \link{lpdf$optcg}(tol,epoch) do optimization with respect to coefficients
#' via conjugate gradient
#' @field \link{lpdf$optnewton}() do optimization via matrix inversion, one Newton
#' step
#' @field lpdf$updateom() update based on recent version of \code{\link{outermod}}
#' @field lpdf$updatepara(para) update using new model parameters
#' @field lpdf$updateterms(terms) update using new \code{terms}
#' @field lpdf$hess() returns the hessian with respect
#' to coefficients
#' @field lpdf$hessgradhyp() returns gradient of \code{hess()} with respect to
#' covariance hyperparameters
#' @field lpdf$hessgradpara() returns the gradient of \code{hess()} with respect to
#' model parameters
#' @field lpdf$diaghess() returns the diagonal of the hessian with
#' respect to coefficients
#' @field lpdf$diaghessgradhyp() returns the gradient of \code{diaghess()} with
#' respect to covariance hyperparameters
#' @field lpdf$diaghessgradpara() returns the gradient of \code{diaghess()} with
#' respect to model parameters
#' @field lpdf$paralpdf(para) compute the log-prior on the parameters, useful for
#' fitting
#' @field lpdf$paralpdf_grad(para) gradient of \code{paralpdf(para)}
#' @seealso container class: \code{\link{lpdfvec}}
#' @seealso derived classes: \code{\link{loglik_std}},
#' \code{\link{loglik_gauss}}, \code{\link{loglik_gda}},
#' \code{\link{logpr_gauss}}
#' @returns no returns, this is a class which contains methods
NULL
#' @name lpdfvec
#' @aliases
#' Rcpp_lpdfvec-class Rcpp_lpdfvec
#' @title Vector of \code{lpdf} instances
#' @description
#' \preformatted{
#' logpdf = new(lpdfvec, loglik, logpr)
#' }
#' This is a class where each instance contains two \code{\link{lpdf}}
#' instances and can be
#' manipulated as a single instance. It presumes both are based on the same
#' \code{\link{outermod}} instance, thus they share hyperparameters. However
#' the model parameters are concatenated. Currently also includes variations
#' on marginal adjustments.
#'
#' Currently it is designed only for a pair, but the ordering is arbitrary.
#'
#' @param loglik one reference to a \code{lpdf} instance
#' @param logpr another reference to a \code{lpdf} instance that shares
#' \code{\link{outermod}} with \code{loglik}
#' @field lpdfvec$domarg A boolean that controls if marginal adjustment is
#' done
#' @seealso base class: \code{\link{lpdf}}
#' @returns no returns, this is a class which contains methods
NULL
#' @name lpdf$optcg
#' @title Optimization via Conjugate Gradient
#' @description
#' \preformatted{
#' lpdf$optcg(tol,epoch)
#' }
#' This optimizes the coefficient vector \code{coeff} using conjugate gradient.
#' It currently is designed only for quadratic \code{\link{lpdf}} instances.
#' @param tol A positive double representing tolerance, default is
#' \code{0.001}.
#' @param epoch A positive integer representing the maximum number of steps
#' conjugate gradient will take.
#' @seealso \code{\link{lpdf}}
#' @returns nothing is returned, the class instance is updated
NULL
#' @name lpdf$optnewton
#' @title Optimization via Newton's Method
#' @description
#' \preformatted{
#' lpdf$optnewton()
#' }
#' This optimizes the coefficient vector \code{coeff} using Newton's Method.
#' It currently is designed only for quadratic \code{\link{lpdf}} instances.
#' It should take a single step.
#' @seealso \code{\link{lpdf}}
#' @returns nothing is returned, the class instance is updated
NULL
#' @name loglik_std
#' @aliases
#' Rcpp_loglik_std-class Rcpp_loglik_std
#' @title Gaussian errors
#' @description
#' \preformatted{
#' loglik = new(loglik_std, om, terms, y, x)
#' }
#' This is a standard model which has the form
#' \deqn{y = \langle \phi(x), \theta \rangle + \varepsilon, \varepsilon \sim
#' N(0,\sigma^2)}
#' where \eqn{\phi(x)} is the basis, \eqn{\theta} is the coefficient vector,
#' \eqn{\varepsilon} is an unseen noise vector.
#' The parameter vector is of length 1 where
#' \code{para} \eqn{= \log(\sigma)}. It is a slower (sometimes)
#' version of \code{\link{loglik_gauss}} but allows for complete marginal
#' inference.
#' @param om an \code{\link{outermod}} instance to be referred to
#' @param terms a matrix of \code{terms}, must have as many columns as dims in
#' \code{om}
#' @param y a vector of observations
#' @param x a matrix of predictors, must have as many columns as dims in
#' \code{om} and the same number of rows as \code{y}
#' @inherit lpdf description
#' @seealso base class: \code{\link{lpdf}}
#' @returns no returns, this is a class which contains methods
NULL
#' @name loglik_gauss
#' @aliases
#' Rcpp_loglik_gauss-class Rcpp_loglik_gauss
#' @title Gaussian errors, large scale
#' @description
#' \preformatted{
#' loglik = new(loglik_gauss, om, terms, y, x)
#' }
#' This is a standard model which has the form
#' \deqn{y = \langle \phi(x), \theta \rangle + \varepsilon, \varepsilon \sim
#' N(0,\sigma^2)}
#' where \eqn{\phi(x)} is the basis, \eqn{\theta} is the coefficient vector,
#' \eqn{\varepsilon} is an unseen noise vector.
#' The parameter vector is of length 1 where
#' \code{para} \eqn{= \log(\sigma)}. It is a faster (sometimes) version of
#' \code{\link{loglik_std}} but can only handle diagonal variational
#' inference.
#' @param om an \code{\link{outermod}} instance to be referred to
#' @param terms a matrix of \code{terms}, must have as many columns as dims in
#' \code{om}
#' @param y a vector of observations
#' @param x a matrix of predictors, must have as many columns as dims in
#' \code{om} and the same number of rows as \code{y}
#' @inherit lpdf description
#' @seealso base class: \code{\link{lpdf}}
#' @returns no returns, this is a class which contains methods
NULL
#' @name loglik_gda
#' @aliases
#' Rcpp_loglik_gda-class Rcpp_loglik_gda
#' @title Gaussian errors with diagonal adjustment
#' @description
#' \preformatted{
#' loglik = new(loglik_gda, om, terms, y, x)
#' }
#' This is a standard model which has the form
#' \deqn{y = \langle \phi(x), \theta \rangle + \delta(x) + \varepsilon,
#' \delta(x) \sim N(0, \lambda g(x)), \varepsilon \sim N(0,\sigma^2)}
#' where \eqn{\phi(x)} is the basis, \eqn{\theta} is the coefficient vector,
#' \eqn{\delta(x)} is unseen vector corresponding to unmodeled
#' variance \eqn{\lambda g(x)}, \eqn{\varepsilon} is an unseen noise vector.
#' The parameter vector is of length 2 where
#' \eqn{\sigma=} \code{exp(para[0])} and \eqn{\lambda=}\code{exp(2*para[1])}.
#' @param om an \code{\link{outermod}} instance to be referred to
#' @param terms a matrix of \code{terms}, must have as many columns as dims in
#' \code{om}
#' @param y a vector of observations
#' @param x a matrix of predictors, must have as many columns as dims in
#' \code{om} and the same number of rows as \code{y}
#' @inherit lpdf description
#' @seealso base class: \code{\link{lpdf}}
#' @returns no returns, this is a class which contains methods
NULL
#' @name logpr_gauss
#' @aliases
#' Rcpp_logpr_gauss-class Rcpp_logpr_gauss
#' @title Gaussian prior
#' @description
#' \preformatted{
#' logpr = new(logpr_gauss, om, terms)
#' }
#' This is a standard model of coefficients which has them as drawn
#' independently
#' from
#' \deqn{ \theta_i \sim N(0, \rho c_i)}
#' where \eqn{c_i} is the variance supplied by \code{om} for the \eqn{i}th term.
#' The parameter vector is of length 1 where
#' \eqn{\rho=} \code{exp(para[0])}.
#' @param om an \code{\link{outermod}} instance to be referred to
#' @param terms a matrix of \code{terms}, must have as many columns as dims in
#' \code{om}
#' @inherit lpdf description
#' @seealso base class: \code{\link{lpdf}}
#' @returns no returns, this is a class which contains methods
NULL
#' @name covf
#' @aliases
#' Rcpp_covf-class Rcpp_covf
#' @title covariance function class
#' @description This is a base class designed to handle the specific features of
#' covariances needed for outerbase. Polymorphism allows for the implied
#' methods to be used across several similar classes.
#' @field covf$hyp hyperparameters for this specific correlation function
#' @field covf$lowbnd,covf$uppbnd upper and lower bounds for the inputs to the
#' covariance function.
#' @field covf$cov(x1,x2) returns the covariance matrix between two vectors of
#' inputs \code{x1} and \code{x2}
#' @field covf$covdiag(x1) returns the diagonal of the covariance matrix between
#' \code{x1} and itself
#' @field covf$cov_gradhyp(x1,x2) returns a cube of the gradient the \code{cov}
#' with respect to the covariance hyperparameters
#' @seealso derived class: \code{\link{covf_mat25}},
#' \code{\link{covf_mat25pow}},
#' \code{\link{covf_mat25ang}}
#' @returns no returns, this is a class which contains methods
NULL
#' @name covf_mat25
#' @aliases
#' Rcpp_covf_mat25-class Rcpp_covf_mat25
#' @title Matern covariance function
#' @description
#' \preformatted{
#' covf = new(covf_mat25)
#' }
#' This is the standard Matern covariance function which has form
#' \deqn{c(x_1,x_2) = (1+ |h| + h^2/3) \exp(-|h|) }
#' where \eqn{h = (x_1-x_2)/\rho} and \eqn{\rho}=\code{exp(2*hyp[0])}.
#' @seealso base class: \code{\link{covf}}
#' @returns no returns, this is a class which contains methods
NULL
#' @name covf_mat25pow
#' @aliases
#' Rcpp_covf_mat25pow-class Rcpp_covf_mat25pow
#' @title Matern covariance function with power transform
#' @description
#' \preformatted{
#' covf = new(covf_mat25pow)
#' }
#' This is the standard Matern covariance function with a power transformation
#' which has form
#' \deqn{c(x_1,x_2) = (1+ |h| + h^2/3) \exp(-|h|) }
#' where \eqn{h = (x_1^\alpha-x_2^\alpha)/\rho} and \code{hyp} is a two
#' dimensional vector with \eqn{\rho}=\code{exp(2*hyp[0]+0.25*hyp[1])}
#' and \eqn{\alpha}=\code{exp(0.25*hyp[1])}.
#' @seealso base class: \code{\link{covf}}
#' @returns no returns, this is a class which contains methods
NULL
#' @name covf_mat25ang
#' @aliases
#' Rcpp_covf_mat25ang-class Rcpp_covf_mat25ang
#' @title Matern covariance function with angular transform
#' @description
#' \preformatted{
#' covf = new(covf_mat25ang)
#' }
#' This is the standard Matern covariance function with a power transformation
#' which has form
#' \deqn{c(x_1,x_2) = (1+ |h| + h^2/3) \exp(-|h|) }
#' where \deqn{h = \sqrt{(\sin(x_1)-\sin(x_2))^2/\rho_s +
#' (\cos(x_1)-\cos(x_2))^2/\rho_c}.}
#' \code{hyp} is a two dimensional vector with
#' \eqn{\rho_s}=\code{exp(2*hyp[0])} and \eqn{\rho_c}=\code{exp(2*hyp[1])}.
#' @seealso base class: \code{\link{covf}}
#' @returns no returns, this is a class which contains methods
NULL
#' @name predictor
#' @aliases
#' Rcpp_predictor-class Rcpp_predictor
#' @title prediction class
#' @description
#' \preformatted{
#' pred = new(predictor, loglik)
#' }
#' This is a base class design to allow for coherent building of
#' predictions across multiple models. Unlike many base classes in this
#' package, it is meant to be directly used.
#'
#' @param loglik An \code{\link{lpdf}} instance, specifically that starts with
#' \code{loglik}, to build the predictor
#' @field predictor$update(x) update the current input to \code{x} for prediction
#' @field predictor$mean() return the vector of means for the prediction
#' @field predictor$var() return the vector of variances for the prediction
#' @field predictor$setnthreads(k) specifics \code{k} as the number of threads
#' to use
#' @returns no returns, this is a class which contains methods
NULL
Try the outerbase package in your browser
Any scripts or data that you put into this service are public.
outerbase documentation built on June 9, 2022, 5:08 p.m.
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
# Generated by using Rcpp::compileAttributes() -> do not edit by hand
# Generator token: 10BE3573-1514-4C36-9D1C-5A225CD40393
#' @name setcovfs
#' @title Set covariance functions
#' @description
#' \preformatted{
#' setcovfs(om, covnames)
#' }
#' Sets the covariance functions for an outermod class instance.
#' This is first thing one does when creating an outermod instance.
#' @param om an \code{\link{outermod}} instance
#' @param covnames a vector of strings of the covariance functions
#' @examples
#' om = new(outermod)
#' setcovfs(om, c("mat25", "mat25", "mat25"))
#' setcovfs(om, c("mat25", "mat25pow", "mat25", "mat25ang"))
#' @seealso \code{\link{outermod}}
#' @returns no value is returned, \code{om} is updated
NULL
#' @name setknot
#' @title Set knot points
#' @description
#' \preformatted{
#' setknot(om, knotslist)
#' }
#' Sets the knot points of \code{om} to \code{knotslist} to estimate the
#' eigenfunctions and eigenvalues. It will naturally check if the knot points
#' have the same dimension as the covariance functions. It will also check if
#' the knot points are within reasonable bounds for the covariance functions.
#' @param om an \code{\link{outermod}} instance
#' @param knotslist a list of one dimensional vectors
#' @examples
#' om = new(outermod)
#' setcovfs(om, c("mat25", "mat25", "mat25"))
#' knotslist = list(seq(0,1,by=0.01),seq(0,1,by=0.01),seq(0,1,by=0.01))
#' setknot(om, knotslist)
#' @seealso \code{\link{outermod}}, \code{\link{setcovfs}}
#' @returns no value is returned, \code{om} is updated
NULL
#' @name gethyp
#' @title Get the hyperparameters
#' @description
#' \preformatted{
#' hyp = gethyp(om)
#' }
#' Gets the current hyperparameters from an \code{\link{outermod}} instance. It
#' formats them in a way that makes reading in \code{R} easier.
#' @param om an \code{\link{outermod}} instance
#' @examples
#' om = new(outermod)
#' setcovfs(om, c("mat25", "mat25", "mat25"))
#' hyp = gethyp(om)
#' print(hyp)
#' @returns a vector of parameters
#' @seealso \code{\link{outermod}}
NULL
#' @name getpara
#' @title Get the model parameters
#' @description
#' \preformatted{
#' para = getpara(logpdf)
#' }
#' This function gets the current parameters from an \code{\link{lpdf}} class
#' instance. It formats them in a way that makes reading in \code{R} easier.
#' @returns a vector of parameters
#' @param logpdf an \code{\link{lpdf}} class instance
NULL
#' @name outermod
#' @aliases
#' Rcpp_outermod-class Rcpp_outermod
#' @title Outer product-type model
#' @description This is a class used to construct \code{\link{outerbase}}
#' class instances. It stores key information for constructing a basis.
#' @field \link{outermod$updatehyp}(hyp) update hyperparameters
#' @field \link{outermod$selectterms}(numterms) find best \code{numterms} terms
#' @field \link{outermod$getvar}(terms) find variances of coefficients
#' associated with \code{terms}
#' @examples
#' om = new(outermod)
#' setcovfs(om, c("mat25", "mat25", "mat25"))
#' setknot(om,
#' list(seq(0,1,by=0.01),seq(0,1,by=0.01),seq(0,1,by=0.01)))
#' terms = om$selectterms(40)
#' coeffvar =om$getvar(terms)
#' hyp = gethyp(om)
#' hyp[1:2] = 0.5
#' om$updatehyp(hyp)
#' coeffvar = om$getvar(terms)
#' @seealso \code{\link{outerbase}} the main product from an outermod
#' @seealso \code{\link{setcovfs}}, \code{\link{setknot}},
#' \code{\link{gethyp}}
#' @returns no returns, this is a class which contains methods
NULL
#' @name outermod$updatehyp
#' @title Update hyperparameters
#' @description
#' \preformatted{
#' outermod$updatehyp(hyp)
#' }
#' Updates the hyperparameters for the instance of outermod.
#' @param hyp A vector of hyperparameters
#' @returns no value is returned, the class instance is updated
#' @seealso \code{\link{outermod}}
NULL
#' @name outermod$selectterms
#' @title Select optimal terms
#' @description
#' \preformatted{
#' terms = om$selectterms(numterms)
#' }
#' Returns the best \code{numterms} given \code{outermod} currently using
#' maximum variance criteria.
#' @param numterms number of basis \code{terms} desired
#' @returns a matrix of \code{terms}
#' @seealso \code{\link{outermod}}
NULL
#' @name outermod$getvar
#' @title Get variance of coefficients
#' @description
#' \preformatted{
#' coeffvar = outermod$getvar(terms)
#' }
#' Returns the variance of the coefficients associated with \code{terms}.
#' @param terms a matrix of \code{terms}
#' @returns a vector of variances of each coefficient
#' @seealso \code{\link{outermod}}
NULL
#' @name outerbase
#' @aliases
#' Rcpp_outerbase-class Rcpp_outerbase
#' @title Outer product-type basis
#' @description
#' \preformatted{
#' ob = new(outerbase, om, x)
#' }
#' Class that handles the basis for a given set of points
#' \code{x}.
#' @param x a matrix of predictors, must have as many columns as dims in
#' \code{om}
#' @field nthreads number of threads for \code{omp} to use
#' @field \link{outerbase$getbase}(k) to get each dimensions basis
#' functions
#' @field \link{outerbase$getmat}(terms) to get the basis matrix at
#' \code{terms}
#' @field \link{outerbase$build}() to (re)build the basis instance
#' @field \link{outerbase$matmul}(terms,a) matrix multiply without
#'building the basis matrix
#' @field \link{outerbase$tmatmul}(terms,a) transpose matrix multiply
#' without building the basis matrix
#' @examples
#' om = new(outermod)
#' setcovfs(om, c("mat25", "mat25", "mat25"))
#' setknot(om,
#' list(seq(0,1,by=0.025),seq(0,1,by=0.025),seq(0,1,by=0.025)))
#' x = matrix(runif(10*3),ncol=3)
#' ob = new(outerbase, om, x)
#' terms = om$selectterms(40)
#' basismat = ob$getmat(terms)
#' @seealso \code{\link{outermod}} the core element that controls outerbase
#' @returns no returns, this is a class which contains methods
NULL
#' @name outerbase$getbase
#' @title Get base functions
#' @description
#' \preformatted{
#' basis_func = outerbase$getbase(k)
#' }
#' Returns the basis for dimension \code{k}. Designed mostly for
#' visualization.
#' @param k An integer from that corresponds to the dimension.
#' @returns a matrix of evaluated basis functions
#' @seealso \code{\link{outerbase}}
NULL
#' @name outerbase$getmat
#' @title Get basis matrix
#' @description
#' \preformatted{
#' basismat = outerbase$getmat(terms)
#' }
#' Returns the basis matrix for a given set of \code{terms}.
#' @param terms a matrix of terms
#' @returns a matrix of evaluated basis functions based on
#' \code{terms}.
#' @seealso \code{\link{outerbase}}
NULL
#' @name outerbase$build
#' @title Builds the outerbase
#' @description
#' \preformatted{
#' outerbase$build()
#' }
#' Build (or re-build) a basis based on the recent evaluation
#' of \code{\link{outermod}}.
#' @seealso \code{\link{outerbase}}
#' @returns nothing is returned, the class instance is updated
NULL
#' @name outerbase$matmul
#' @title Matrix multiply
#' @description
#' \preformatted{
#' b = outerbase$matmul(terms, a)
#' }
#' Multiplies the basis times a vector without building the basis
#' matrix.
#' @param terms a matrix of \code{terms}
#' @param a a vector of length the same as the rows in \code{terms}
#' @returns a vector resulting from the matrix multiplication
#' @seealso \code{\link{outerbase}}
NULL
#' @name outerbase$tmatmul
#' @title Transpose Matrix multiply
#' @description
#' \preformatted{
#' b = outerbase$tmatmul(terms, a)
#' }
#' Multiplies the transpose of the basis times a vector without
#' building the basis matrix.
#' @param terms a matrix of \code{terms}
#' @param a a vector of length the same as the rows in \code{outerbase}
#' @returns a vector resulting from the matrix multiplication
#' @seealso \code{\link{outerbase}}
NULL
#' @name lpdf
#' @aliases
#' Rcpp_lpdf-class Rcpp_lpdf
#' @title Log probability density function class
#' @description This is a base class designed to handle the learning of
#' the underlying coefficients, hyperparameters, and parameters associated with
#' a specific learning instance. Polymorphism allows for the implied methods to
#' be used across several similar classes.
#'
#' @field lpdf$val current value
#' @field lpdf$para current model parameters
#' @field lpdf$coeff current coefficients
#' @field lpdf$compute_val on calling \code{update}, compute value and store in
#' \code{val}
#' @field lpdf$grad current gradient with respect to coefficients
#' @field lpdf$gradhyp current gradient with respect to covariance hyperparameters
#' @field lpdf$gradpara current gradient with respect to model parameters
#' @field lpdf$compute_grad on calling \code{update}, compute gradient with
#' respect to coefficients and store in \code{grad}
#' @field lpdf$compute_gradhyp on calling \code{update}, compute gradient
#' with respect to covariance hyperparameters and store in \code{gradhyp}
#' @field lpdf$compute_gradpara on calling \code{update}, compute gradient
#' with respect to model parameters and store in \code{gradpara}
#' @field lpdf$update(coeff) update using new coefficients
#' @field \link{lpdf$optcg}(tol,epoch) do optimization with respect to coefficients
#' via conjugate gradient
#' @field \link{lpdf$optnewton}() do optimization via matrix inversion, one Newton
#' step
#' @field lpdf$updateom() update based on recent version of \code{\link{outermod}}
#' @field lpdf$updatepara(para) update using new model parameters
#' @field lpdf$updateterms(terms) update using new \code{terms}
#' @field lpdf$hess() returns the hessian with respect
#' to coefficients
#' @field lpdf$hessgradhyp() returns gradient of \code{hess()} with respect to
#' covariance hyperparameters
#' @field lpdf$hessgradpara() returns the gradient of \code{hess()} with respect to
#' model parameters
#' @field lpdf$diaghess() returns the diagonal of the hessian with
#' respect to coefficients
#' @field lpdf$diaghessgradhyp() returns the gradient of \code{diaghess()} with
#' respect to covariance hyperparameters
#' @field lpdf$diaghessgradpara() returns the gradient of \code{diaghess()} with
#' respect to model parameters
#' @field lpdf$paralpdf(para) compute the log-prior on the parameters, useful for
#' fitting
#' @field lpdf$paralpdf_grad(para) gradient of \code{paralpdf(para)}
#' @seealso container class: \code{\link{lpdfvec}}
#' @seealso derived classes: \code{\link{loglik_std}},
#' \code{\link{loglik_gauss}}, \code{\link{loglik_gda}},
#' \code{\link{logpr_gauss}}
#' @returns no returns, this is a class which contains methods
NULL
#' @name lpdfvec
#' @aliases
#' Rcpp_lpdfvec-class Rcpp_lpdfvec
#' @title Vector of \code{lpdf} instances
#' @description
#' \preformatted{
#' logpdf = new(lpdfvec, loglik, logpr)
#' }
#' This is a class where each instance contains two \code{\link{lpdf}}
#' instances and can be
#' manipulated as a single instance. It presumes both are based on the same
#' \code{\link{outermod}} instance, thus they share hyperparameters. However
#' the model parameters are concatenated. Currently also includes variations
#' on marginal adjustments.
#'
#' Currently it is designed only for a pair, but the ordering is arbitrary.
#'
#' @param loglik one reference to a \code{lpdf} instance
#' @param logpr another reference to a \code{lpdf} instance that shares
#' \code{\link{outermod}} with \code{loglik}
#' @field lpdfvec$domarg A boolean that controls if marginal adjustment is
#' done
#' @seealso base class: \code{\link{lpdf}}
#' @returns no returns, this is a class which contains methods
NULL
#' @name lpdf$optcg
#' @title Optimization via Conjugate Gradient
#' @description
#' \preformatted{
#' lpdf$optcg(tol,epoch)
#' }
#' This optimizes the coefficient vector \code{coeff} using conjugate gradient.
#' It currently is designed only for quadratic \code{\link{lpdf}} instances.
#' @param tol A positive double representing tolerance, default is
#' \code{0.001}.
#' @param epoch A positive integer representing the maximum number of steps
#' conjugate gradient will take.
#' @seealso \code{\link{lpdf}}
#' @returns nothing is returned, the class instance is updated
NULL
#' @name lpdf$optnewton
#' @title Optimization via Newton's Method
#' @description
#' \preformatted{
#' lpdf$optnewton()
#' }
#' This optimizes the coefficient vector \code{coeff} using Newton's Method.
#' It currently is designed only for quadratic \code{\link{lpdf}} instances.
#' It should take a single step.
#' @seealso \code{\link{lpdf}}
#' @returns nothing is returned, the class instance is updated
NULL
#' @name loglik_std
#' @aliases
#' Rcpp_loglik_std-class Rcpp_loglik_std
#' @title Gaussian errors
#' @description
#' \preformatted{
#' loglik = new(loglik_std, om, terms, y, x)
#' }
#' This is a standard model which has the form
#' \deqn{y = \langle \phi(x), \theta \rangle + \varepsilon, \varepsilon \sim
#' N(0,\sigma^2)}
#' where \eqn{\phi(x)} is the basis, \eqn{\theta} is the coefficient vector,
#' \eqn{\varepsilon} is an unseen noise vector.
#' The parameter vector is of length 1 where
#' \code{para} \eqn{= \log(\sigma)}. It is a slower (sometimes)
#' version of \code{\link{loglik_gauss}} but allows for complete marginal
#' inference.
#' @param om an \code{\link{outermod}} instance to be referred to
#' @param terms a matrix of \code{terms}, must have as many columns as dims in
#' \code{om}
#' @param y a vector of observations
#' @param x a matrix of predictors, must have as many columns as dims in
#' \code{om} and the same number of rows as \code{y}
#' @inherit lpdf description
#' @seealso base class: \code{\link{lpdf}}
#' @returns no returns, this is a class which contains methods
NULL
#' @name loglik_gauss
#' @aliases
#' Rcpp_loglik_gauss-class Rcpp_loglik_gauss
#' @title Gaussian errors, large scale
#' @description
#' \preformatted{
#' loglik = new(loglik_gauss, om, terms, y, x)
#' }
#' This is a standard model which has the form
#' \deqn{y = \langle \phi(x), \theta \rangle + \varepsilon, \varepsilon \sim
#' N(0,\sigma^2)}
#' where \eqn{\phi(x)} is the basis, \eqn{\theta} is the coefficient vector,
#' \eqn{\varepsilon} is an unseen noise vector.
#' The parameter vector is of length 1 where
#' \code{para} \eqn{= \log(\sigma)}. It is a faster (sometimes) version of
#' \code{\link{loglik_std}} but can only handle diagonal variational
#' inference.
#' @param om an \code{\link{outermod}} instance to be referred to
#' @param terms a matrix of \code{terms}, must have as many columns as dims in
#' \code{om}
#' @param y a vector of observations
#' @param x a matrix of predictors, must have as many columns as dims in
#' \code{om} and the same number of rows as \code{y}
#' @inherit lpdf description
#' @seealso base class: \code{\link{lpdf}}
#' @returns no returns, this is a class which contains methods
NULL
#' @name loglik_gda
#' @aliases
#' Rcpp_loglik_gda-class Rcpp_loglik_gda
#' @title Gaussian errors with diagonal adjustment
#' @description
#' \preformatted{
#' loglik = new(loglik_gda, om, terms, y, x)
#' }
#' This is a standard model which has the form
#' \deqn{y = \langle \phi(x), \theta \rangle + \delta(x) + \varepsilon,
#' \delta(x) \sim N(0, \lambda g(x)), \varepsilon \sim N(0,\sigma^2)}
#' where \eqn{\phi(x)} is the basis, \eqn{\theta} is the coefficient vector,
#' \eqn{\delta(x)} is unseen vector corresponding to unmodeled
#' variance \eqn{\lambda g(x)}, \eqn{\varepsilon} is an unseen noise vector.
#' The parameter vector is of length 2 where
#' \eqn{\sigma=} \code{exp(para[0])} and \eqn{\lambda=}\code{exp(2*para[1])}.
#' @param om an \code{\link{outermod}} instance to be referred to
#' @param terms a matrix of \code{terms}, must have as many columns as dims in
#' \code{om}
#' @param y a vector of observations
#' @param x a matrix of predictors, must have as many columns as dims in
#' \code{om} and the same number of rows as \code{y}
#' @inherit lpdf description
#' @seealso base class: \code{\link{lpdf}}
#' @returns no returns, this is a class which contains methods
NULL
#' @name logpr_gauss
#' @aliases
#' Rcpp_logpr_gauss-class Rcpp_logpr_gauss
#' @title Gaussian prior
#' @description
#' \preformatted{
#' logpr = new(logpr_gauss, om, terms)
#' }
#' This is a standard model of coefficients which has them as drawn
#' independently
#' from
#' \deqn{ \theta_i \sim N(0, \rho c_i)}
#' where \eqn{c_i} is the variance supplied by \code{om} for the \eqn{i}th term.
#' The parameter vector is of length 1 where
#' \eqn{\rho=} \code{exp(para[0])}.
#' @param om an \code{\link{outermod}} instance to be referred to
#' @param terms a matrix of \code{terms}, must have as many columns as dims in
#' \code{om}
#' @inherit lpdf description
#' @seealso base class: \code{\link{lpdf}}
#' @returns no returns, this is a class which contains methods
NULL
#' @name covf
#' @aliases
#' Rcpp_covf-class Rcpp_covf
#' @title covariance function class
#' @description This is a base class designed to handle the specific features of
#' covariances needed for outerbase. Polymorphism allows for the implied
#' methods to be used across several similar classes.
#' @field covf$hyp hyperparameters for this specific correlation function
#' @field covf$lowbnd,covf$uppbnd upper and lower bounds for the inputs to the
#' covariance function.
#' @field covf$cov(x1,x2) returns the covariance matrix between two vectors of
#' inputs \code{x1} and \code{x2}
#' @field covf$covdiag(x1) returns the diagonal of the covariance matrix between
#' \code{x1} and itself
#' @field covf$cov_gradhyp(x1,x2) returns a cube of the gradient the \code{cov}
#' with respect to the covariance hyperparameters
#' @seealso derived class: \code{\link{covf_mat25}},
#' \code{\link{covf_mat25pow}},
#' \code{\link{covf_mat25ang}}
#' @returns no returns, this is a class which contains methods
NULL
#' @name covf_mat25
#' @aliases
#' Rcpp_covf_mat25-class Rcpp_covf_mat25
#' @title Matern covariance function
#' @description
#' \preformatted{
#' covf = new(covf_mat25)
#' }
#' This is the standard Matern covariance function which has form
#' \deqn{c(x_1,x_2) = (1+ |h| + h^2/3) \exp(-|h|) }
#' where \eqn{h = (x_1-x_2)/\rho} and \eqn{\rho}=\code{exp(2*hyp[0])}.
#' @seealso base class: \code{\link{covf}}
#' @returns no returns, this is a class which contains methods
NULL
#' @name covf_mat25pow
#' @aliases
#' Rcpp_covf_mat25pow-class Rcpp_covf_mat25pow
#' @title Matern covariance function with power transform
#' @description
#' \preformatted{
#' covf = new(covf_mat25pow)
#' }
#' This is the standard Matern covariance function with a power transformation
#' which has form
#' \deqn{c(x_1,x_2) = (1+ |h| + h^2/3) \exp(-|h|) }
#' where \eqn{h = (x_1^\alpha-x_2^\alpha)/\rho} and \code{hyp} is a two
#' dimensional vector with \eqn{\rho}=\code{exp(2*hyp[0]+0.25*hyp[1])}
#' and \eqn{\alpha}=\code{exp(0.25*hyp[1])}.
#' @seealso base class: \code{\link{covf}}
#' @returns no returns, this is a class which contains methods
NULL
#' @name covf_mat25ang
#' @aliases
#' Rcpp_covf_mat25ang-class Rcpp_covf_mat25ang
#' @title Matern covariance function with angular transform
#' @description
#' \preformatted{
#' covf = new(covf_mat25ang)
#' }
#' This is the standard Matern covariance function with a power transformation
#' which has form
#' \deqn{c(x_1,x_2) = (1+ |h| + h^2/3) \exp(-|h|) }
#' where \deqn{h = \sqrt{(\sin(x_1)-\sin(x_2))^2/\rho_s +
#' (\cos(x_1)-\cos(x_2))^2/\rho_c}.}
#' \code{hyp} is a two dimensional vector with
#' \eqn{\rho_s}=\code{exp(2*hyp[0])} and \eqn{\rho_c}=\code{exp(2*hyp[1])}.
#' @seealso base class: \code{\link{covf}}
#' @returns no returns, this is a class which contains methods
NULL
#' @name predictor
#' @aliases
#' Rcpp_predictor-class Rcpp_predictor
#' @title prediction class
#' @description
#' \preformatted{
#' pred = new(predictor, loglik)
#' }
#' This is a base class design to allow for coherent building of
#' predictions across multiple models. Unlike many base classes in this
#' package, it is meant to be directly used.
#'
#' @param loglik An \code{\link{lpdf}} instance, specifically that starts with
#' \code{loglik}, to build the predictor
#' @field predictor$update(x) update the current input to \code{x} for prediction
#' @field predictor$mean() return the vector of means for the prediction
#' @field predictor$var() return the vector of variances for the prediction
#' @field predictor$setnthreads(k) specifics \code{k} as the number of threads
#' to use
#' @returns no returns, this is a class which contains methods
NULL
Try the outerbase package in your browser
Any scripts or data that you put into this service are public.
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
Embedding an R snippet on your website
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.