[SciPy-dev] Another GSoC idea

[David Warde-Farley]

On 25-Mar-09, at 5:56 PM, Robert Kern wrote:

> However, once one gets enough
> distributions to be interesting and to give assurances that one has
> covered all of the use cases in the design, the student can forgo the
> remaining distributions to work on a way to combine these distribution
> objects into probabilistic models that can be converted to efficient
> MCMC, EM, or other such numerical codes for estimation.

That is indeed a compelling idea.

In what little experimentation I've done with PyMC2, it seems to be  
quite well designed and it would definitely be a mistake to ignore it  
in crafting such a project, it would at the very least be a good  
choice as of the downstream paths for model rendering. I can't attest  
to how well it copes with being scaffolding for other inference  
methods, but it would be great to have all of the MCMC infrastructure  
right there to test against, for example, a variational approximation  
to the posterior that you've written code for (or, in the future we  
are pondering, _generated_ code for).

David