HistOptimal

Automatic choice of bin number in regular histogram construction

Description

Choose automatically the number of bin of a regular histogram using penalized likelihood procedure following paper of Birgé-Rozenholc in ESAIM P&S (2006), 10, pp.24-45 "How many bins should be put in a regular histogram?"

• MATLAB release: MATLAB 7 (R14)
• Other requirements: no requirement

[F Moisy]

The file "ReadMe" is unreadable (wrong format?)
I assume everything was explained in this file, because the comments in HistOptimal.m do not help much (no example). The help refers to the files hist_optimal_doc.dvi and hist_optimal_doc.ps, which apparently are also missing in the submission.

[Yves Rozenholc]

I just upload a modified version which will be disponible in 1 or 2 days. Actually, the references to hist_optimal_doc.ps are wrong. The ReadMe is a plain text file that you could open with any text editor.

[F Moisy]

I rate here the new version of the submission -- Although the core of the function (determination of the optimal number of bins for an histogram) is original and interesting, the two other functions (binning of the data and plot of the histogram) are redundant with Matlab's built-in functions.

[Dimitri Shvorob]

The procedure seems to boil down to a simple adjustment, made in 2-3 lines of histoptimal.m. Would it be quite so hard to leave a PDF with the formula and brief explanation? Or leave a link to your paper on ESAIM? 95% of the code, indeed, replicates Matlab's built-in functionality.
 

[Yves Rozenholc]

The hist.m function in matlab do not create a histogram in a statistical meaning. This is one reason to re-construct it with bin surfaces equal to probabilities.
The link to the paper is :
http://www.esaim-ps.org/index.php?option=article&access=standard&Itemid=129&url=/articles/ps/abs/2006/01/ps0322/ps0322.html

[Dimitri Shvorob]

I am not sure I follow. In what sense does HIST 'not create a histogram in a statistical meaning'? And what do you mean by 'bin surfaces equal to probabilities'? If you are referring to binwidths, these are constant, and a hardly equal to any probabilities.