Future Version planning

I believe that OpenBayes has reached a certain state of maturity that a new version should be planified. Lot's of people have contributed to the code, either by helping to fix bugs or by contributing code themselves. We thank all the users for their interest in OpenBayes, which is what makes us continue this project.

So now is the time to put all this together and make a new step forwards...

Version 0.2

Version 0.2 will emphasize on creating a stable base for future development. This is important because it will condition the speed of development in the future. I prefer having 1 or 2 features less and a stable platform, instead of having lot's of features and then debugging the whole thing for years. So in this version it is probable to see some changes in the names of the variables and classes (in order to gain in consistency and homogeneity). The other main point of this version will be to incorporate all the features that were contributed during the last few months and making all the features work transparently for discrete, gaussian and hybrid networks. We insist on this last point because once this is done, any other distribution can be considered (such as SoftMax, NoisyOr, ...) and as long as the specifications are satisfied, we are guarantied that all the inference and learning engines will also work with the new distributions.

Version 0.2 will include :

• Incorporate and test already existing features:
  

• EM Learning

• SEM Learning

• Augmented Bayesian Networks (Dirichlet's a priori for learning)
• MCMC sampling engine with Importance Sampling
• Gaussian Distributions (some modifications and testing)
• ... (anybody wrote something and kept it for himself ? :-) )

• Entirely New features

• Gaussian Potentials: CG, SG potentials that will allow for exact inference and learning in gaussian and hybrid networks
• Discrete Gaussians Distribution (does anybody need this?)
• Sum of Gaussians Distribution
  

• Programming Issues

• Use standard coding for variable and class names
• Clear up the code and inspect the class hierarchy
• Create UML diagrams of OpenBayes (help needed for this part)
• Create more tests on large networks  + a general test sequence after installing (easy contributions)

• Documentation

• Add some new Tutorials (anybody can contribute with this)
• Add lot's of examples (anybody can contribute with this)
• Inspect and enhance the pydocs (developpers only)
• Tutorial for creating a new distribution type (this will help users contribute more easily)
  

When ?

This new version should be available by April 2006.

How to contribute?

Anybody that wishes to contribute or suggest something can post a message on the Forum under the discussion Feature Requests, or you can directly contact Kosta Gaitanis : gaitanis@tele.ucl.ac.be

(It is better to post because other people can also get involved in the discussion. Only mail me for private discussions)

Also, there is a Skype meeting planified for the end of February for those who are interested in actively contributing to the project. Contact me directly for more info.

Future Versions :

A few words concerning version 0.3 which is planified probably for next year, depending on the debugging process, user contributions and most of all : available time!

Ver 0.3 will include

• Dynamic Bayesian Networks
• Basic inference for DBNs
• ...