Re: Binning of continuous data


Tim De Meyer wrote:
Hello,
I have a stretch of continuous data which I need to bin. Although I think it's possible to come up with some heuristics, I'm now looking for a rigid evaluation criterion, preferably based on information theory. The general idea is that, when signals in the data occur, they are binned in one single bin. Number of bins and size of each bin don't matter. For example take the following stretch of data:

1 0 1 0 0 1 0 10 33 22 22 0 1 0 1 13 25 22 0 1 0

The highest evaluation score should then be obtained for(bin edge represented by "|"):
|1 0 1 0 0 1 0 |10 33 22 22| 0 1 0 1 |13 25 22 |0 1 0|

I think this problem has been tackled before so references or formal names for this problem are very welcome.

Note: I'm only searching for an evaluation criterion, not for an algorithm itselve.

Thanks a lot!
Tim

Are you really searching for binning criteria or a method to
tell when the data have switched from one "state" (for lack
of a better term) to the other? If the latter, you might want
to check out something like "median runs test".

Cheers,
Russell

.