[gmx-developers] New statistics mini-library

[Mark Abraham]

David van der Spoel wrote:
> Hi,
> 
> I've implemented a small library with some statistics functions 
> (line-fitting, correlation coefficient, histograms and a bit more), 
> based on old code that was shattered around the analysis tools, but with 
> a cleaner interface, and with a test program. The idea is to extend the 
> library with more general statistics functions as may be needed (e.g. 
> stuff from g_analyze).
> 
> The  question I have is whether this should be made part of the analysis 
> library or the gmx library, and whether this should be stored in a 
> separate directory. For clarity and simplicity, I would propose putting 
> it in a separate directory in the gmxlib directory, as it is not 
> impossible that such functionality may be needed from mdrun as well.
> Comments?

Looks like a good idea to me. I would have used/extended such a library 
several times for my own stuff. Re-implementing wheels for each utility 
program has to be the wrong approach. I like the idea of a separate 
directory to try to encourage the separation of groups of code. That 
said, it's been years now, and I still have only the vaguest idea 
whether a particular bit of code might live in a source file in 
src/mdlib, src/gmxlib or src/kernel...

Mark