[SciPy-user] is there any plan to import BNT(bayesian network toolkit) from matlab?
David Warde-Farley
dwf at cs.toronto.edu
Thu Apr 10 17:23:08 EDT 2008
On 10-Apr-08, at 3:26 PM, Karl Young wrote:
> Stefan, that sounds great. After talking with Jarrod and David
> yesterday
> I'm getting a better feel for how the port might fit into the overall
> SciPy picture (initially as part of the learn scikit). One of the
> things
> I thought was great about the toolkit was Kevin Murphy's overarching
> view of graphical models, e.g. Hidden Markov Models are a particular
> case of his general scheme. But David made the important point
> yesterday
> that if you wanted to use HMM's for something like speech processing
> using such a general approach would be inefficient and it would be
> better to use more specific code (currently existing in SciPy I
> think).
> So one of the first tasks, consistent with what you describe, is to
> generate a priority list for the port, perhaps looking for overlapping
> functionality in SciPy and leaving that stuff out of the initial port.
> I'll start on that (and we can try to reconcile that with what anyone
> else comes up with) and I guess I can post my thoughts on the priority
> list on the wiki. I think I should leave decisions about the interface
> to those more expert in that (I'll be happy to start coding once those
> decisions are made though). What is the title of Chris Bishop's book ?
I believe he's referring to "Pattern Recognition and Machine
Learning", which was published in 2006. It's quickly become a
favourite in many circles, and it contains one of the most
comprehensive treatments of graphical models that I know of in any
textbook. The website for the book: http://research.microsoft.com/
~cmbishop/PRML/
Conveniently enough for all interested, the graphical models chapter
is available as a free sample chapter!
http://research.microsoft.com/~cmbishop/PRML/Bishop-PRML-sample.pdf
Stefan, I think that Bishop's chapters on graphical models, Markov
Chain Monte Carlo, and variational methods are an excellent place to
start.
Off the top of my head, these papers are rather illuminating as well:
http://www.cs.toronto.edu/~roweis/papers/NC110201.pdf
http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=910572
http://www.cs.toronto.edu/~radford/em.abstract.html
Once I'm done my coursework for the term and have more time on my
hands I'll be contributing whatever I can to the effort.
Cheers,
David
More information about the SciPy-User
mailing list