Quick memory win w/ `edgelist()`

[benthestatistician]

For many purposes ISMs can be treated like numeric vectors. Two meaningless examples:

is(big_ism_pairs)
## [1] "BlockedInfinitySparseMatrix" "InfinitySparseMatrix"
## [3] "vector"
head(big_ism_pairs)
## [1] 0.0081 0.5048 0.3906 0.6252 0.6794 0.3088
head(big_ism_pairs *2)
## [1] 0.0162 1.0097 0.7812 1.2504 1.3588 0.6175

Do the dist columns of objects returned by edgelist() follow this pattern? If so then we can save on memory by simplifying (in R/edgelist.R ; branch i54-hinting) "dist = x@.Data" to "dist = x":

load(file="gurm_match_big_ism_pairs.RData")
object_size(big_ism_pairs)
## 584 MB
object_size(big_ism_pairs, optmatch:::edgelist(big_ism_pairs))
## 1.16 GB
edgelist_  <- function(ism) data.frame(i=ism@rownames[ism@rows], 
                                            j=ism@colnames[ism@cols], dist = ism)
object_size(big_ism_pairs, edgelist_(big_ism_pairs))
## 875 MB

(Of course the task also calls for tests to document whether/how these dist variables could be persuaded to adopt the needed behavior.)

[josherrickson]

Per an offline discussion, I used tracemem against big_ism_pairs and it was not triggered by either edgelist or edgelist_.