BTW, factor graphs are generally more popular than Bayes Nets these days, at least in research circles. (Bayesian networks can be seen as a special case of factor graph.)
Those are a lot of great links! But I disagree with your "BTW" line:
I don't think factor graphs are more popular than Bayes nets (a.k.a. directed graphical models), at least not in general machine learning (though perhaps in some particular subfield?). Directed graphs are usually most appropriate for generative models, which are very popular, especially in Bayesian approaches. In graphical models, directed, undirected, and factor graphs are all used in their appropriate contexts.
And it's not accurate to say that directed models (Bayes nets) are a special case of factor graphs: there are conditional independence structures that can be represented by Bayes nets that can't be captured exactly with factor graphs. The canonical example is
Glad you like the links, but I disagree with your disagreement =P
You can use factor graphs for either directed or undirected models. See this paper:
Extending factor graphs so as to unify directed and undirected graphical models
B. J. Frey
Proceedings of the 19th Conference on Uncertainty in Artificial Intelligence, 257-264, Morgan Kaufmann, San Francisco, CA, August 2003.
I'm using Mahout, it does have Bayesian classifiers (Naive Bayes and Complementary Naive Bayes). However, Bayesian algorithms never worked well for my datasets. I went with LDA for my tasks :)
BTW, if you do want to develop a library to work with classifiers, my only "wish" would be to have a good export/import tools. Most libraries i worked with are using their own serialization formats, so it's quite difficult to import/export data without an overhead. If you write several adapters that could allow people to work with datasets through the adapters, i'm quite sure that ecosystem will get created by other people extremely fast. Some people use mongo, some use postgre, others use files, csv, serialized formats. It's great to be able to continue to use your own things without changes when running that kind of a lib.
Other "wish" thing is to have an ability to run classifier as a map/reduce job, and do it easily :)
8 comments
[ 3.1 ms ] story [ 36.3 ms ] threadhttp://search.cpan.org/~kwilliams/AI-Categorizer-0.09/
I'm currently using this (and String::Approx) on a not-so-secret project...
http://compbio.cs.huji.ac.il/FastInf/fastInf/FastInf_Homepag...
http://robotics.stanford.edu/~sgould/svl/
http://code.google.com/p/factorie/
http://people.kyb.tuebingen.mpg.de/jorism/libDAI/
http://research.microsoft.com/en-us/um/cambridge/projects/in...
BTW, factor graphs are generally more popular than Bayes Nets these days, at least in research circles. (Bayesian networks can be seen as a special case of factor graph.)
I don't think factor graphs are more popular than Bayes nets (a.k.a. directed graphical models), at least not in general machine learning (though perhaps in some particular subfield?). Directed graphs are usually most appropriate for generative models, which are very popular, especially in Bayesian approaches. In graphical models, directed, undirected, and factor graphs are all used in their appropriate contexts.
And it's not accurate to say that directed models (Bayes nets) are a special case of factor graphs: there are conditional independence structures that can be represented by Bayes nets that can't be captured exactly with factor graphs. The canonical example is
O --> O <-- O
You can use factor graphs for either directed or undirected models. See this paper:
Extending factor graphs so as to unify directed and undirected graphical models B. J. Frey Proceedings of the 19th Conference on Uncertainty in Artificial Intelligence, 257-264, Morgan Kaufmann, San Francisco, CA, August 2003.
BTW, if you do want to develop a library to work with classifiers, my only "wish" would be to have a good export/import tools. Most libraries i worked with are using their own serialization formats, so it's quite difficult to import/export data without an overhead. If you write several adapters that could allow people to work with datasets through the adapters, i'm quite sure that ecosystem will get created by other people extremely fast. Some people use mongo, some use postgre, others use files, csv, serialized formats. It's great to be able to continue to use your own things without changes when running that kind of a lib.
Other "wish" thing is to have an ability to run classifier as a map/reduce job, and do it easily :)
(LDA is a bit of an overloaded acronym in machine learning, but I'm assuming that's what you mean.)
[1] http://www.cs.princeton.edu/~blei/papers/BleiNgJordan2003.pd...
Thanks for pointing that out, but i was speaking more of concrete algorithm implementation rather than abstracts.
has a pretty good Bayesian library