HISTOGRAM --- A CHALLENGE ---   version 1.2 : Jan. 16, 2016
 version 1.1 : Aug. 24, 2012

Here, we issue a challenge of optimizing a histogram given a set of data points or a series of event times.
We believe that you can write a paper, if you succeed in constructing an algorithm that consistently beats ours.
 
Congratulations! (Jan. 16, 2016)
An algorithm presented by Max Murphy (University of Kansas) beat the Omi-Shinomoto method 14 times out of 20 !
 
(1) Download three sets of event times, which are drawn from hidden underlying rate processes chosen randomly. Event times are not necessarily derived from the Poisson process. The possible methods of drawing event times are described in Ref.[1].
(2) Guess optimal bin sizes for three sets of data so that the histograms best express unknown underlying rate processes.
(3) See your histograms and those determined by the rules of Omi & Shinomoto (2011), Shimazaki & Shinomoto (2007), Scott (1979), and Sturges (1926), by comparison with true underlying rates. These five kinds of histograms will be evaluated in terms of L2 and L1 errors between the histograms and the underlying rate. Your overall ranking will be given according to the average of those rankings.


(1) Download Data.  




(2) Guess optimal bin sizes for three sets of data.

     Data1       Data2       Data3  

(3) Compare the performances.  



















REFERENCES
[1] Omi T. and Shinomoto S. (2011). Optimizing time histograms for non-Poissonian spike trains. Neural Computation 23:3125-3144.
[2] Shimazaki H. and Shinomoto S. (2007). A method for selecting the bin size of a time histogram.
Neural Computation 19:1503-1700.
[3] Scott D.W. (1979). Optimal and data-based histograms. Biometrika, 66, 605–610.
[4] Sturges H.A. (1926). The choice of a class interval. J American Statistical Association: 65–66.

The program was created by Takahiro Omi in collaboration with Shigeru Shinomoto, to whom any questions, comments, or suggestions should be addressed.
Toolbox for analyzing spike data, SULAB