R/feval_spd.R

In mrparker909/MGLMRiem: Multivariate Generalized Linear Models on Riemannian Manifolds

# function E = feval_spd(p,V,X,Y)
# %FEVAL_SPD evaluates the objective function value (the sum of squared geodesic errors) of MGLM on SPD. 
# %
# %    E = FEVAL_SPD(p,V, X, Y)
# %
# %   !! make sure that X is centered if p, V are calculated by centered X !!
#   %
# %   X is a set of column vectors. (dimX X N, where dimX = size(X,1))
# %   p is the base point.
# %   V is a set of symmetric matrices (dim_p x dim_p x dimX).
# %   Y is a set of SPD matrices (dim_p x dim_p x N, where dim_p = size(p,1)).
# %   E is the sum of squared geodesic errors.
# %
# %   See also LOGMAP_SPD, LOGMAP_PT2ARRAY_SPD, EXPMAP_SPD, GSQERR_SPD,
# %   PREDICTION_SPD
# 
# %   Hyunwoo J. Kim
# %   $Revision: 0.1 $  $Date: 2014/06/23 15:42:13 $
#   
#   P_hat = prediction_spd(p,V,X);
# E = gsqerr_spd(Y, P_hat);

#' @title feval_spd
#' @description Calculates the sum of squared geodesic errors (which mglm_spd() tries to minimize).
#' @param X X is a set of column vectors.
#' @param p p is the base point.
#' @param V V is a set of symmetric matrices.
#' @param Y Y is a set of SPD matrices.
#' @export
feval_spd <- function(p,V,X,Y) {
# FEVAL_SPD evaluates the objective function value (the sum of squared geodesic errors) of MGLM on SPD. 
#
#    E = FEVAL_SPD(p, V, X, Y)
#
#   !! make sure that X is centered if p, V are calculated by centered X !!
#
#   X is a set of column vectors. (dimX X N, where dimX = size(X,1))
#   p is the base point.
#   V is a set of symmetric matrices (dim_p x dim_p x dimX).
#   Y is a set of SPD matrices (dim_p x dim_p x N, where dim_p = size(p,1)).
#   E is the sum of squared geodesic errors.
#
#   See also LOGMAP_SPD, LOGMAP_PT2ARRAY_SPD, EXPMAP_SPD, GSQERR_SPD,
#   PREDICTION_SPD

#   Hyunwoo J. Kim
#   $Revision: 0.1 $  $Date: 2014/06/23 15:42:13 $

#   Migrated to R by Matthew RP Parker
#   $Revision: 0.2 $  $Date: 2019/06/06 $  

  P_hat <- prediction_spd(p,V,X)
  E     <- gsqerr_spd(X = Y, X_hat = P_hat)
  return(E)
}