[R] ...... looping

[Martin Maechler]

>>>>> "BaRow" == Barry Rowlingson <B.Rowlingson at lancaster.ac.uk>
>>>>>     on Wed, 25 Oct 2006 11:06:06 +0100 writes:

    BaRow> Xiaofan Cao wrote:
    >> Hi there,
    >> 
    >> I'm writing a program in R that has a few nested loops. I'd like to 
    >> monitor the progress when the program is running and be able to estimate 
    >> the remaining time.

    BaRow> A long time ago I started writing some code to give R something like 
    BaRow> an 'iterator' object. You could do this:

    >> ml=loop(5)

    >> while(iterate(ml))
    BaRow> + {cat("doing ",iteration(ml)," of ",N(ml),"\n","Ending at 
    BaRow> ",predictEnd(ml),"\n");sleep(5)}

    BaRow> doing  1  of  5
    BaRow> Ending at  Wed 25 Oct 2006 11:00:05 BST
    BaRow> doing  2  of  5
    BaRow> Ending at  Wed 25 Oct 2006 11:00:20 BST
    BaRow> doing  3  of  5
    BaRow> Ending at  Wed 25 Oct 2006 11:00:20 BST
    BaRow> doing  4  of  5
    BaRow> Ending at  Wed 25 Oct 2006 11:00:20 BST
    BaRow> doing  5  of  5
    BaRow> Ending at  Wed 25 Oct 2006 11:00:20 BST

    BaRow> you use loop(N) to construct a 1:N loop object, while(iterate(ml)) to 
    BaRow> loop round it, iteration(ml) to get the current iteration number, N(ml) 
    BaRow> to get the iteration limit, and predictEnd(ml) to guess when the whole 
    BaRow> thing will finish.

    BaRow> All the information about the loop is encapsulated in the ml object.

    BaRow> It needs a chunk of polishing up and nobody seemed that interested in 
    BaRow> it last time I mentioned it. My particular application was to MCMC, 
    BaRow> where you could have an MCMC iterator object that was a subclass of my 
    BaRow> simple loop class, and then you could have methods like if(isBurnIn(ml)) 
    BaRow> to decide when to start taking samples, or if(!isThinned(ml)) to decide 
    BaRow> whether to store a sample from a thinned chain. Again, all the info 
    BaRow> encapsulated in the loop object.

    BaRow> Another advantage is that unlike for(i in 1:10000000) it doesn't 
    BaRow> create a vector of 10000000 objects...

    BaRow> If anyone thinks this is worth me working on then I'll try and find 
    BaRow> some spare time (hah!) to fix it up. Or if anyone wants to take over, I 
    BaRow> can throw my code at you at see if it sticks.

I think it's very well worth pursuing and "publishing" one way
or the other. IIRC, even with R-core (Luke?) people have had related ideas
(not mainly for MCMC, but for remedying the fact that you
 don't want to construct 1:10000000 or its version with another '0'),
even thinking about using (simple versions of these) for
"standard R".

Probably the topic then really belongs to R-devel rather than
R-help.

Martin Maechler, ETH Zurich