| Histogram Bug [message #28128] |
Mon, 26 November 2001 08:43 |
John-David T. Smith
Messages: 384
Registered: January 2000
|
Senior Member |
|
|
Here's another ugly IDL-crashing bug:
IDL> print,histogram([1,2],NBINS=1)
It happens in both 5.4 and 5.5. The problem is with 1 bin. It vanishes
if BINSIZE is specified, except, of course, as:
IDL> print,histogram([1,2],NBINS=1,BINSIZE=0)
Not that this last one is meaningful, but still it should err
gracefully. In fact any small enough binsize will cause the crash.
A workaround is to use a large enough nonzero BINSIZE for any histogram
call for which NBINS could ever end up being unity. In reality,
histogram(a,NBINS=1) should equal n_elements(a), and if you don't set
your BINSIZE appropriately, you could miss values.
This all relates, I suspect, to the somewhat strange relation assumed by
histogram() for BINSIZE, given NBINS:
BINSIZE = (MAX-MIN)/(NBINS-1)
with the side effect of *changing* any input MAX, instead of the more
sensible (in my opinion), and max-preserving:
BINSIZE = (MAX-MIN)/NBINS
This strange behavior we discussed long ago, with the conclusion that
it's probably too late to change. Here's a snippet from my original
complaint:
> I'm not sure why I didn't notice this before. The recently introduced
> NBINS keyword to histogram is incorrectly defined.
>
> BINSIZE
> Set this keyword to the size of the bin to
> use. If this keyword is not specified, and
> NBINS is not set, then a bin size of 1
> is used. If NBINS is set, the default is
> BINSIZE = (MAX � MIN) / (NBINS � 1).
>
> For example, consider 3 bins over the range [0,1]. Should be a binsize
> of 1/3, yes? Instead, you get a binsize of 1/2, and bins from:
>
> [0,.5]
> [.5,1.]
> [1.,1.5]
JD
P.S. Interestingly, the obvious failure case NBINS=0 is explicitly
detected.
|
|
|
|
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
