Thread Subject: local maxima of a histogram

   Is there a MATLAB function to calculate all the local
maxima of a histogram?

    Otherwise, I think I`ll have to calculate the mode,
remove (or zero) that value, then calculate it again, down
to some specified level.

   I have to find all the maxima, then test to see if any
of the maxima are within a given value of the others, then
do some more combining.

           Thanks, Alan

James wrote:
> Is there a MATLAB function to calculate all the local
> maxima of a histogram?
>
> Otherwise, I think I`ll have to calculate the mode,
> remove (or zero) that value, then calculate it again, down
> to some specified level.
>
> I have to find all the maxima, then test to see if any
> of the maxima are within a given value of the others, then
> do some more combining.
>
> Thanks, Alan

Not sure I completely understand what you need, but assuming that you
are dealing with an m x n matrix of values that represent your
histogram data, you can use the 'max' command to tell MATLAB to find
the max along a particular dimension (see 'help max' for more). One of
many solutions would be to use 'repmat' and 'diff' to do the second
part of your analysis.

James,

Try using histc, and then sorting the output.

Hope this helps,
Rob S.

"James " <jalanNOSPAMthomas@verizon.net> wrote in message
<fltp79$4l8$1@fred.mathworks.com>...
> Is there a MATLAB function to calculate all the local
> maxima of a histogram?
>
> Otherwise, I think I`ll have to calculate the mode,
> remove (or zero) that value, then calculate it again,
down
> to some specified level.
>
> I have to find all the maxima, then test to see if any
> of the maxima are within a given value of the others,
then
> do some more combining.
>
> Thanks, Alan
>
>
>

More specifically...

numpoints = 1000;
data = floor(randn(numpoints,1)*100)+500;
counts = histc(data,1:numpoints);
[freq,value] = sort(counts,'descend');

So then the most frequent value will be value(1), which
occurs freq(1) times. The next most frequent value is value
(2) which occurs freq(2) times. Etc.

Cheers,
Rob S.

"Rob S" <rslazas.dontspamme@gmail.com> wrote in message
<fltqpv$h38$1@fred.mathworks.com>...
> James,
>
> Try using histc, and then sorting the output.
>
> Hope this helps,
> Rob S.
>
> "James " <jalanNOSPAMthomas@verizon.net> wrote in message
> <fltp79$4l8$1@fred.mathworks.com>...
> > Is there a MATLAB function to calculate all the
local
> > maxima of a histogram?
> >
> > Otherwise, I think I`ll have to calculate the mode,
> > remove (or zero) that value, then calculate it again,
> down
> > to some specified level.
> >
> > I have to find all the maxima, then test to see if
any
> > of the maxima are within a given value of the others,
> then
> > do some more combining.
> >
> > Thanks, Alan
> >
> >
> >
>

"Rob S" <rslazas.dontspamme@gmail.com> wrote in message
<fltsi8$4g4$1@fred.mathworks.com>...
> More specifically...
>
> numpoints = 1000;
> data = floor(randn(numpoints,1)*100)+500;
> counts = histc(data,1:numpoints);
> [freq,value] = sort(counts,'descend');
>
> So then the most frequent value will be value(1), which
> occurs freq(1) times. The next most frequent value is value
> (2) which occurs freq(2) times. Etc.
>
> Cheers,
> Rob S.
>
> "Rob S" <rslazas.dontspamme@gmail.com> wrote in message
> <fltqpv$h38$1@fred.mathworks.com>...
> > James,
> >
> > Try using histc, and then sorting the output.
> >
> > Hope this helps,
> > Rob S.

SNIP

Hi
The above process does not give only extremes, but a
sequence of the largest class frequencies. The given problem
solves the function

function [Lx,Xh] = mnxtr(hist)
% Extremes of histogram
% hist (m,n) matrix class frequencies
% Lx logical matrix of peaks of hist
% Xh cell array of extremes in hist rows
[m,n] = size(hist);
Lx = false(m,n);
Xh = {};
for k = 1:m
    Lhl = extr(hist(k,:));
    Lx(k,:) = Lhl{1}';
    Xh = [Xh; {hist(Lx(k,:))}];
end

Try
m=10; n=15;
hst = round(rand(m,n)*100) % matrix of class frequencies
[Lx,Xh]=mnxtr(hist)
Xh{2} % extremes in the second row of hst

Hope it helps

Mira

Function extr is from FEX, Id. 10272

Mira

James wrote:
> Is there a MATLAB function to calculate all the local
> maxima of a histogram?
>
> Otherwise, I think I`ll have to calculate the mode,
> remove (or zero) that value, then calculate it again, down
> to some specified level.
>
> I have to find all the maxima, then test to see if any
> of the maxima are within a given value of the others, then
> do some more combining.

It's worth noting that in many situations the local maxima in a
histogram will be very dependent on the bin that you use. So unless you
have some a priori reason for choosing a specific set of bins (e.g., you
have integer data), you might think about using something like KSDENSITY
from the Statistics Toolbox instead of a histogram. It does kernel
density estimation, which is, roughly, like a smoothed histogram, and
does not depend on choosing a particular set of bins.

Hope this helps.

- Peter Perkins
   The MathWorks, Inc.