hist.m can handle only one column when nbins > 29

[John W. Eaton]

On  6-Mar-2008, Robert S. Mahurin wrote:

| When hist.m is asked to produce a histogram with 30 or more bins, it
| uses a "more efficient" algorithm whos implementation contains a size
| error.

| Beginning hist.m line 107:
| 
|   else
|     ## The following algorithm works fastest for n greater than about 30.
|     ## Put cutoff elements between boundaries, integrate over all
|     ## elements, keep totals at boundaries.
| ### [s, idx] = sort ([y; cutoff]);
|     [s, idx] = sort ([y; repmat(cutoff, 1, columns (y))]);
|     len = rows (y);
|     chist = cumsum (idx <= len);
|     t1 = zeros (1, columns (y));
| ### t2 = reshape (chist(idx > len), size (cutoff));
|     t2 = reshape (chist(idx > len), rows (cutoff), columns (y));
|     t3 = chist(end,:) - sum (isnan (y));
|     chist = [t1; t2; t3];
|   endif
| 
| Also, eliminating the temporaries t1,t2,t3 seems to reduce memory
| usage by eliminating some irrelevant matrix reference.  If I replace
| the code between the two chist assignments with
| 
|     chist = cumsum (idx <= len);
|     chist = [zeros(1, columns (y)); \
| 	     reshape(chist(idx > len), rows (cutoff), columns (y)); \
| 	     chist(end,:) - sum(isnan (y));]; 

I made the following changes.

Thanks,

jwe


-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: diffs
Url: https://www.cae.wisc.edu/pipermail/bug-octave/attachments/20080306/07061d78/attachment-0001.ksh