On Jan 21, 2013, at 1:42 PM, "Miller, Timothy" 
<[email protected]> wrote:
> I can't answer conclusively, but I believe those are all models trained using 
> the clearTK framework.  There may be a way of packaging them as separate 
> files rather than jars but I'm not sure if that would have any benefit since:
> 1) You almost always will not be/should not be modifying models trained using 
> machine learning methods,
> 2) It may clutter up the models directory -- if you have a 10 class 
> classifier it will build 10 one-vs-all classifiers and package them together 
> (along with some metadata) and I think it is valuable to encapsulate this 
> even for many of the core ctakes developers.
> 
> As far as whether these models change or evolve, they will certainly change 
> as features or more data are added, but I wouldn't really say evolve, at 
> least not in the sense that svn can take advantage of.  A small change in the 
> training input will change these models almost completely so it wouldn't 
> really be valuable to, e.g., view diffs between component model files even if 
> they are ascii.
> 
> Dima or Steve Bethard may be able to answer better but I'm not sure if they 
> are checking email today.

I couldn't have said it any better. Tim was right on there.

Steve

Reply via email to