vecpack: vecpack: a package for packing things into a vector

In dahtah/vecpack: Helps Turn Objects and Lists into Vectors, to Facilitate Interaction with Optimisers and Samplers

Description

vecpack is an experimental package for R, meant to help you stuff various things into a single vector (which is easy), and get them back in the right shape (which isn't). It lets you interface with optim in a more natural way: in the following example we compute a low-rank + diagonal approximation to a matrix. The most natural way of writing the cost function involves two arguments: a matrix and a scalar. Using vecpack you can send that cost function directly to optim.

Author(s)

Simon Barthelme

Examples

#Packing things into a vector is easy
list(A=matrix(0,2,2),b=1:3) %>% sapply(as.vector) %>% do.call("c",.)
#Going the other way isn't: many objects correspond to the same vector
#gen.vunpack generates an "unpacking" function from an example
l <- list(A=matrix(0,2,2),b=1:3)
vunpack <- gen.vunpack(l)
vpack(l)
vpack(l) %>% vunpack

#The most common use for vpack is for interfacing with optimisation packages
#In this example we compute a rank-2 + diagonal approximation 
#to the matrix X
X <- cor(USArrests ) 
#A cost function over two parameters: a is a scalar and B is a matrix
cfun <- function(a,B)
   {
       lowrank <- a*diag(4) + tcrossprod(B)
       sum((lowrank-X)^2)
   }
guess <- list(a=0,B=matrix(0,4,2))
#vpoptim wraps optim
res <- vpoptim(guess,cfun)

dahtah/vecpack documentation built on May 14, 2019, 3:27 p.m.