R devel - Dec 2025 - Speed question: passing arguments vs environment

On 2025-12-02 7:41 a.m., Therneau, Terry M., Ph.D. via R-devel
wrote:
> I have a complex likelihood function f() to maximize, with lots of
arguments (some of which set up indexes for derivatives, for instance).
> When using something like optim(), one can pass these arguments through via
its ? arg, or could make the likelihood function f() live in the same
environment as the main routine so they are found directly.    Is there any
advantage of one versus the other wrt speed?    At the end of the day, f() may
get called thousands of times in a Hamiltonian MCMC.
> 
> Since R does not replicate arguments that are used in a read-only fashion,
one might expect little to no penalty for having them on the call chain, unless
the bookkeeping for copy-on-write is itself time consuming.
By the way, a nice way to put the args in the environment of the 
objective function is to use local() or a builder, e.g.

objective <- local({
   arg1 <- 1
   arg2 <- 2
   arg3 <- 3
   function(x) {
   # objective code here that can see arg1, arg2, arg3
   }
})

or

makeObjective <- function(arg1, arg2, arg3) {
   force(arg1) # evaluate the promises
   force(arg2)
   force(arg3)

   function(x) {
     # objective code here that can see arg1, arg2, arg3
   }
}

objective <- makeObjective(1,2,3)

Duncan Murdoch

Duncan's suggestion to time things is important -- and would make a very
useful short communication or blog!
There are frequently differences of orders of magnitude in timing.

I'll also suggest that it is worth some crude timings of different solvers.
There is sufficient variation
over problems that this won't decide definitively which solver is fastest,
but you might eliminate one or
two that are poor for your situation. Depending on numbers of parameters,
I'd guess ncg or its predecessor
Rcgmin will be relatively good. LBFGS variants can be good, but sometimes seem
to toss up disasters. Most
of these can be accessed with optimx package to save coding. By removing some
checks and safeguards in
optimx you could likely speed up things a bit too.

If full optimum is not needed, some attention to early stopping might be
worthwhile, but I've seen lots
of silly mistakes made playing with tolerances, and if you go that route, choose
a custom termination
rule that fits your particular problem or you'll get rubbish.

JN

On 2025-12-02 08:48, Duncan Murdoch wrote:
> On 2025-12-02 7:41 a.m., Therneau, Terry M., Ph.D. via R-devel wrote:
>> I have a complex likelihood function f() to maximize, with lots of
arguments (some of which set up indexes for
>> derivatives, for instance).
>> When using something like optim(), one can pass these arguments through
via its ? arg, or could make the likelihood
>> function f() live in the same environment as the main routine so they
are found directly.??? Is there any advantage of
>> one versus the other wrt speed???? At the end of the day, f() may get
called thousands of times in a Hamiltonian MCMC.
>>
>> Since R does not replicate arguments that are used in a read-only
fashion, one might expect little to no penalty for
>> having them on the call chain, unless the bookkeeping for copy-on-write
is itself time consuming.
> 
> By the way, a nice way to put the args in the environment of the objective
function is to use local() or a builder, e.g.
> 
> objective <- local({
>  ? arg1 <- 1
>  ? arg2 <- 2
>  ? arg3 <- 3
>  ? function(x) {
>  ? # objective code here that can see arg1, arg2, arg3
>  ? }
> })
> 
> or
> 
> makeObjective <- function(arg1, arg2, arg3) {
>  ? force(arg1) # evaluate the promises
>  ? force(arg2)
>  ? force(arg3)
> 
>  ? function(x) {
>  ??? # objective code here that can see arg1, arg2, arg3
>  ? }
> }
> 
> objective <- makeObjective(1,2,3)
> 
> Duncan Murdoch
> 
> ______________________________________________
> R-devel at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-devel