| setting histogram bin sizes? [message #61608] |
Tue, 22 July 2008 08:52  |
jeffnettles4870
Messages: 111
Registered: October 2006
|
Senior Member |
|
|
Hi folks,
I'm looking for suggestions for a way to set bin sizes for a histogram
when I don't know much about the data before calculating the
histogram. Here's my situation: I'm putting together some code that
takes a hyperspectral image cube and extracts a series of one-band
parameters from the cube (band depth at a certain wavelength, etc.).
In trying to assess which of these parameters is most useful for our
particular application i thought about calculating a histogram for
each parameter. The problem is that these parameter images (one band,
floating point images per parameter) will not necessarily fall into
the same range. Many have possible values of 0 - 1, but they won't
necessarily take up that entire range. Some however, will not have
possible values of 0 - 1, but could instead have numbers in the 10s or
even hundreds. Some parameters have values that are actually in log
space.
I know that I could simply set the NBINS keyword to HISTOGRAM(), but
then the question would become how many bins to use? I did some quick
searching, and there are a few attempts at calculating bin sizes or
the number of bins on Wikipedia (http://en.wikipedia.org/wiki/
Histogram). Short of any other information, i am going to use an
equation from that page that is at least based on the standard
deviation of the data. But, since I don't have a lot to go on, I
would very much like to have input from anyone on this newsgroup who
might have any suggestions for me.
Thanks,
Jeff
|
|
|
|
| Re: setting histogram bin sizes? [message #61702 is a reply to message #61608] |
Sat, 26 July 2008 12:33  |
dasergatskov
Messages: 1
Registered: July 2008
|
Junior Member |
|
|
On Jul 22, 10:52 am, "Jeff N." <jeffnettles4...@gmail.com> wrote:
> Hi folks,
>
> I'm looking for suggestions for a way to set bin sizes for a histogram
> when I don't know much about the data before calculating the
> histogram. Here's my situation: I'm putting together some code that
> takes a hyperspectral image cube and extracts a series of one-band
> parameters from the cube (band depth at a certain wavelength, etc.).
> In trying to assess which of these parameters is most useful for our
> particular application i thought about calculating a histogram for
> each parameter. The problem is that these parameter images (one band,
> floating point images per parameter) will not necessarily fall into
> the same range. Many have possible values of 0 - 1, but they won't
> necessarily take up that entire range. Some however, will not have
> possible values of 0 - 1, but could instead have numbers in the 10s or
> even hundreds. Some parameters have values that are actually in log
> space.
>
> I know that I could simply set the NBINS keyword to HISTOGRAM(), but
> then the question would become how many bins to use? I did some quick
> searching, and there are a few attempts at calculating bin sizes or
> the number of bins on Wikipedia (http://en.wikipedia.org/wiki/
> Histogram). Short of any other information, i am going to use an
> equation from that page that is at least based on the standard
> deviation of the data. But, since I don't have a lot to go on, I
> would very much like to have input from anyone on this newsgroup who
> might have any suggestions for me.
>
> Thanks,
> Jeff
I found Kevin Knuth (http://www.huginn.com/knuth/) paper
Optimal Data-Based Binning for Histograms
http://arxiv.org/abs/physics/0605197
to be quite useful.
Sincerely,
Dmitri.
--
|
|
|
|
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
