Core Hunter 3 documentation

[Pietro Delfino]

Good morning,

Thanks for developing the program.

My question is about the Core Hunter 3 documentation.

The algorithms behind the program are the same of Core Hunter 2?

I read that for Core Hunter 3 the JAMES framework has been used; is it the only difference with previous versions?

And then, should we cite the same Core Hunter 1 and 2 original papers if using version 3 ?

Thanks!

[Herman De Beukelaer]

Dear Pietro,

The main changes in Core Hunter 3 are:

1. Improved distance-based evaluation measures (E-NE and A-NE) as explained on the homepage.
2. Support for more data types (various marker formats, phenotypic traits, precomputed distance matrix).
3. Reimplemented from scratch using the JAMES framework.
4. Default optimization algorithm is again the parallel tempering algorithm (a.k.a. REMC) as in Core Hunter 1.
   The Mixed Replica search from Core Hunter 2 is no longer used.
   

We switched back to parallel tempering since the more complex Mixed Replica search used by Core Hunter 2 is no longer

needed due to the improved distance-based evaluation measures, that are more easily optimized with local searches as

compared to minimum distance (which was one of the objectives in Core Hunter 2). Core Hunter 3 also provides a fast

mode in which a basic stochastic hill-climber is used (referred to as plain local search in the Core Hunter 2 paper, and

as random descent in JAMES and Core Hunter 3).

You can find some more documentation in the R package. For example, have a look at

> ?sampleCore

A new paper with more details is on its way. For now, please cite the original papers if you use Core Hunter 3 for your research.

If you have any more questions, let us know.

Best regards, 

Herman

Op donderdag 2 maart 2017 09:17:00 UTC+1 schreef Pietro Delfino:

[Pietro Delfino]

Thanks a lot, very clear!

Pietro