Rray generating functions

[juangomezduaso]

I think it could be useful to generate rrays from a function over its indexes. To illustrate and brainstorm this idea I have made some alternatives for you to taste it.

library(rray)
#> Warning: package 'rray' was built under R version 3.5.3
# A user friendly version:
rray_formula <- function(dimen, formula){
  names<- names(dimen)
  stopifnot(!is.null(names) )
  e <- rlang::enquo(formula)
  seqs <- purrr::map(as.list(dimen), function(i)1:i)
  frame <- expand.grid(seqs)
  cells<- rlang::eval_tidy(e, data = frame)
  rray(cells, dimen)
}
(rray_formula(c(a=2,b=3),a+b^2) )
#> <rray<dbl>[,3][6]>
#>      [,1] [,2] [,3]
#> [1,]    2    5   10
#> [2,]    3    6   11
(rray_formula(c(x=2,y=3),ifelse(x >= y, 1,0)) )
#> <rray<dbl>[,3][6]>
#>      [,1] [,2] [,3]
#> [1,]    1    0    0
#> [2,]    1    1    0

# A sligthly more verbose alternative:
rray_pmap <- function(dimen,fun){
  seqs <- purrr::map(as.list(dimen), function(i)1:i)
  frame <- expand.grid(seqs)
  l <- as.list(frame)
  names(l)<-NULL
  cells <- purrr::pmap_dbl(l,fun)
  rray(cells, dimen)
}
(rray_pmap(c(2,3), function(x,y)x+y^2) )
#> <rray<dbl>[,3][6]>
#>      [,1] [,2] [,3]
#> [1,]    2    5   10
#> [2,]    3    6   11
(rray_pmap(c(2,3),function(x,y)if(x >= y)1 else 0) )
#> <rray<dbl>[,3][6]>
#>      [,1] [,2] [,3]
#> [1,]    1    0    0
#> [2,]    1    1    0

# A different one, where all indexes go  as a single vector:
rray_vmap <- function(dimen,fun){
  seqs <- purrr::map(as.list(dimen), function(i)1:i)
  matrix <- t(expand.grid(seqs))
  l <- purrr::array_branch(matrix,margin=2)
  cells <- purrr::map_dbl(l,fun)
  rray(cells, dimen)
}
(rray_vmap(c(2,3), function(v) v[1]+v[2]^2) )
#> <rray<dbl>[,3][6]>
#>      [,1] [,2] [,3]
#> [1,]    2    5   10
#> [2,]    3    6   11
(rray_vmap(c(2,3),function(v)if(v[1] >= v[2])1 else 0) )
#> <rray<dbl>[,3][6]>
#>      [,1] [,2] [,3]
#> [1,]    1    0    0
#> [2,]    1    1    0

# The last one is less apealing, but far more useful for programming
# As an example, here is a quick rray_shift() implementation based on it:

rray_shift <-function(x,axis,n){
  rray_vmap(dim(x), function(v) 
    if(v[axis]>n){
      l<-as.list(v)
      l[[axis]]<-l[[axis]]-n
      do.call(`[`,c(list(x),l))
    } else NA)
}
rr=rray(1:6,2:3)  
( rray_shift(rr,2,1)  )
#> <rray<dbl>[,3][6]>
#>      [,1] [,2] [,3]
#> [1,]   NA    1    3
#> [2,]   NA    2    4

^{Created on 2019-06-12 by the reprex package (v0.2.1)}

[juangomezduaso]

An interesting (but perhaps too weird) alternative to these functions would be to have an S3 class of "virtual rrays", consisting of just the function and probaby the dimensionality , but not any concrete dimension. These "vrrays" would operate between themselves (virtually, to yield another vrray having , for example , (A+B)$fun = function(v) A$fun(v) +B$fun(v). And would operate with (a)rrays instantiating a compatible rray with the functions above (with broadcasting included) and yielding a rray as the result. Other functions like reshape, transpose, etc would also be easy to define, returning modified virtual rrays . OTOH, subseting would return a "real" rray. Functions like rray_yank(), rray_sum() etc would have no sense in these "infinite rrays" . Not the case of dim(x) which could, obviously, return: rep(Inf, rray_dims(x) ) ;)