00:17:51 Stanislav Kolenikov: TIL: lobstr::obj_size() -- I have been working with some rpart objects that show up as having size of 500Kb in RStudio environment pane / evaluated by object.size() , but actually eating up 10Gb as shown by the RStudio memory tracker. I suspect they are hauling the environment with them (which is a data frame with 10M rows so I can see how it can take a Gb or two) without really showing it properly.
00:26:55 Jon Harmon (jonthegeek): force <- function (x) x
00:27:27 Stas Kolenikov (NORC): arg_match() ?
00:30:37 Stas Kolenikov (NORC): this is probably a @Jon question -- is there a way to implement <<- with assign(..., envir=whatever) ?
00:30:54 Stas Kolenikov (NORC): Replying to "this is probably a @..."
maybe in this simple example of the stateful functions snippet
00:41:01 Jon Harmon (jonthegeek): Replying to "this is probably a @..."
Kinda this?
testing <- function() {
env <- rlang::caller_env()
i <- unlist(unname(mget("i", envir = env, ifnotfound = 0L)))
print(assign("i", i + 1L, pos = env))
}
testing()
I had to make sure it has a value before I start. There are probably other/cleaner ways to do that step.
00:41:58 Stas Kolenikov (NORC): Reacted to "Kinda this?
testing ..." with 👍
00:42:22 Stas Kolenikov (NORC): Replying to "this is probably a @..."
probably. I will check if that works in my context. Thanks!!
00:43:50 Stas Kolenikov (NORC): Replying to "this is probably a @..."
I think I needed this superassignment to happen from one function I owed to another function I owed so I can make sure the variable is defined in the parent.
00:43:59 Jon Harmon (jonthegeek): Replying to "this is probably a @..."
In general, rlang::caller_env() is super useful. It's really just base::parent.frame(n + 1) (where n is the number of callers to go back, default 1), but it's easier to remember IMO.
00:44:27 Stas Kolenikov (NORC): Reacted to "In general, rlang::..." with ‼️
00:45:41 Stas Kolenikov (NORC): Replying to "this is probably a @..."
right, I was trying to figure out in the debug window -- what is it that I need to hook to the pos argument?? It is not explicitly there in the environment's objects so I did not have that language to express my problem.
00:48:06 Victor Ordu: We have the ecdf() function in Base R
00:48:58 Stas Kolenikov (NORC): I wonder if that mechanics is used somewhere in glm() for real, as it takes functions (for links/families) as inputs
00:49:32 Stas Kolenikov (NORC): maybe the multiple imputation package mice as they have to use functions as arguments (or attributes of data frame columns)
00:49:56 Stas Kolenikov (NORC): maybe some machine learning aggregators .. ? I am less familiar with those
00:55:48 Martin Venegas: that's good to know
01:00:54 Stas Kolenikov (NORC): well it looks like there are edge cases where you want to do an expensive computation and keep it in the factory environment, and then use factoried functions to use that