About the first Release

So I finally got around the release thebeast. Despite being a 0.0.1 version there is actually a lot of work in it. I really hope this can be helpful, as it was helpful to me. My view on it is basically: ‘something like a CRF kinda tool, but it allows to you capture a larger class of correlations, not just sequential ones’. However, there are a few catches:

Catch 1: With the expressive power Markov Logic gives you it is clear that there are lot of models you can define but for which inference and/or learning are infeasible. I got a good feeling for what works and what not, but found it hard to put this into words. I will add more and more information regarding this to wiki but for now I’ll have to just let you play around with it and find out yourself (and post your troubles here or on the discuss group).

Catch 2: there are a few subtleties to take in consideration in terms of the order in which commands may be called. Thus the easiest way for you to define your own model would be to use the skeleton code in the example directory.

Catch 3: As is, the code is really hard to extend and maintain for anyone but me, at least on some levels, because I put far too much focus on premature optimization 🙂 They will be future versions (wait for the 1.x.x release line) that change this (possibly slower).

Catch 4: our file format is different from the format alchemy uses. This is pretty annoying I guess and I hope to change in some future version.