[Bioc-devel] R_CheckUserInterrupt guidelines (was Re: R_CheckUserInterrupt in IRanges)

hi, let me take the chance that you're talking about this to check
whether anybody finds anything odd in the following strategy. in the
package i develop (qpgraph - devel version) for the computationally
intensive C-loops i calculate the percentage of progress and each time
this percentage changes in one integer unit i call
R_CheckUserInterrupt() and i also take the chance to process
window-system events with R_ProcessEvents() (if i understand correctly
this is what this function does). here is the code snippet:

ppct=-1;
k=0;

for (i = 0; i < l_int-1; i++) {
?int i2 = INTEGER(pairup_ij_int)[i] - 1;

?for (j = i+1; j < l_int; j++) {
? ?int j2 = INTEGER(pairup_ij_int)[j] - 1;

? ?REAL(nrrMatrix)[i2+j2*n_var] = qp_edge_nrr(REAL(S),n_var,
INTEGER(N)[0], i2, j2, q, INTEGER(nTests)[0],REAL(alpha)[0]);

? ?REAL(nrrMatrix)[j2+i2*n_var] = REAL(nrrMatrix)[i2+j2*n_var];
? ?k++;
? ?pct = (int) ((k * 100) / n_adj);
? ?if (pct != ppct) {
? ? ?if (INTEGER(verbose)[0]) { ## show progress when
? ? ? ?if (pct % 10 == 0) ? ? ? ## verbose=TRUE
? ? ? ? ?Rprintf("%d",pct);
? ? ? ?else
? ? ? ? ?Rprintf(".",pct);
? ? ? ?R_FlushConsole();
? ? ?}
? ? ?R_CheckUserInterrupt();
#ifdef Win32
? ? ?R_ProcessEvents();
#endif
#ifdef HAVE_AQUA
? ? ?R_ProcessEvents();
#endif
? ? ?ppct = pct;
? ?}
?}
}

let me clarify that the double nested loop iterates more or less through
every pair of variables and thus this computation grows quadratically in
the number of input variables in the dataset.

robert.

On Mon, 2010-01-11 at 20:07 -0800, Henrik Bengtsson wrote:

To 2nd what Seth's says, I think it is worth showing an explicit
example of how to check at regular intervals, e.g. ?for (i = 0; i < n;
i++) { if (i %% 100 == 0) R_CheckUserInterrupt(); ... }

/Henrik

On Mon, Jan 11, 2010 at 3:50 PM, Seth Falcon <sfalcon at fhcrc.org> wrote:

On 1/11/10 1:31 PM, Patrick Aboyoun wrote:

I just modified the C or Fortran code section of package guidelines to
reflect this new entry:

http://wiki.fhcrc.org/bioc/Package_Guidelines#c-or-fortran-code

This looks reasonable.

It is worth noting, however, that calling R_CheckUserInterrupt is not
entirely free. ?If you have a tight loop it might make more sense to check
at intervals rather than each time through and the choice of when to check
will depend very much on the computation.

Here's an example using the inline package (code is at the bottom) with a
silly double loop to compare checking vs not:

With R_CheckUserInterrupt inner loop:

system.time(z <- funx(10000L, TRUE))

? user ?system elapsed
?1.310 ? 0.001 ? 1.313

No checking:

system.time(z <- funx(10000L, FALSE))
? user ?system elapsed
? 0.08 ? ?0.00 ? ?0.08

Now this may be completely unrealistic, but I suspect that a sensible take
away message is that we should be making sure our C code that loops does
call R_CheckUserInterrupt, but depending on the computation, not at every
loop iteration.

+ seth


## example code is below.

library("inline")

code <- "
int i, j, sum = 0, int n = INTEGER(max)[0];
int check_inter = LOGICAL(do_check)[0];
SEXP ans;
PROTECT(ans = allocVector(INTSXP, n));
for (i = 0; i < n; i++) {
? ?INTEGER(ans)[i] = i;
? ?for (j = 0; j < n; j++) {
? ? ? ?sum = i + j;
? ? ? ?if (check_inter) {
? ? ? ? ? ?R_CheckUserInterrupt();
? ? ? ?}
? ?}
}
UNPROTECT(1);
return ans;
"

funx <- cfunction(signature(max="integer", do_check="logical"), code)

system.time(z <- funx(10000L, TRUE))
system.time(z <- funx(10000L, FALSE))


--
Seth Falcon
Bioconductor Core Team | FHCRC