Monday, May 12, 2008

Object-Oriented Bayesian Inference II

A while ago, I blogged about a little piece of code I had written as an example for my book.

Back then, it was a fun exercise for me, but it was clearly unfinished (and it still is) but worth as an example of how to model a mathematical object, a Bayesian random variate, as a python class.

Despite its simplicity and incompleteness, that example, still attracts attention, and people contact me asking for the code. In order to better serve those interested in this problem, I decided to create a project on google code to host the code.

I am still interested in taking this code further, extending its applicability to real problems. So if anyone is willing to help, the project site will hopefully be a good channel for that.


DesiLinguist said...

Hi !

I am a Ph.D. student in Computer Science and I work in statistical natural language processing. I have some experience with working on bayesian inference but I am willing to learn and I think this would be a fantastic opportunity to do so. Almost all of my research work is done with python and I am also a contributing developer to the NLTK project (a natural language toolkit written in Python). I would be very interested in getting involved heavily in this project ! Thanks !

boggie said...

Hi Flavio,

I'm now working with a large-scale neuronal network simulator NEST and its Python interface PyNN, although a newbie in Python, but I'm very interested in your project, can I also be part of it?


Unknown said...

Thanks for the interest guys,

please tell me how you would like to contribute, and we can take it from there...


boggie said...

Do u have any specific features that need to be implemented, such that I can also get familiar with the code?


Unknown said...

Hi Jie,

what is most need at this point is to write a test suite, to make sure the the classes at General.Bayes (Continuous and Discrete) work for all distributions provided by scipy.stats. It may be necessary to modify BayesVar class to deal with distributions which take parameters other than "Loc" and "scale".

If you don't feel like working on this and have a suggestion of something you would like to pursue instead, let me know.