[bip] Announcing PEBL (and asking for assistance)

[Abhik Shah]

Hi Bruce,
   Thanks for your comments. Until now, pebl has only been used within
our lab so the installation and documentation haven't gotten much
attention (yet).

Documentation
----------------------
1. The google code site and documentation now link to the actual license.
2. Documentation lists the specific versions of required dependencies
and platforms where pebl is known to run.

Similar Software
-------------------------
OpenBayes and BNT are primarily focused on parameter learning and
inference while pebl focuses solely on structure learning.  The other
packages do support structure learning but it seems like an add-on
rather than core functionality. They don't include all the learning
algorithms and options that pebl does and don't offer any way to run
code in parallel. I agree that having one package for all Bayes net
related learning is a good idea and I'm interested in integrating pebl
with the BayesNet SciKit.

Code
--------
I've uploaded a tarball of Pebl 0.9.8 to the google code site (and
it's also on pypi) and the latest version is always in svn trunk.

I hope I've addressed all your comments.. I really do appreciate the
time you (and others here) are taking to help me with this.

Thanks,
Abhik


On Fri, Jun 20, 2008 at 11:25 AM, Bruce Southey <bsouthey at gmail.com> wrote:
> Abhik Shah wrote:
>>
>> Hi Titus,
>>    Fine, so you want a tutorial that actually teaches rather than
>> something quick that lets me claim I have a tutorial?!?
>>
>> I've created a bug report
>> (http://code.google.com/p/pebl-project/issues/detail?id=16) for this
>> and I've updated the tutorial using an example from a real dataset
>> (Spellman cell cycle) and better explanation about the process and
>> results.
>>
>> Updated docs at: http://ano.malo.us/pebl/docs/
>> Updated tutorial:  http://ano.malo.us/pebl/docs/tutorial.html
>>
>> The explanation for the results is still weak but a screencast might
>> be a better solution.
>>
>> Thanks,
>> Abhik.
>>
>>
>> On Tue, Jun 17, 2008 at 8:19 PM, C. Titus Brown <ctb at msu.edu> wrote:
>>
>>>
>>> On Tue, Jun 17, 2008 at 05:34:41PM -0400, Abhik Shah wrote:
>>> -> Hi,
>>> ->   I've been working on pebl (Python Environment for Bayesian Learning)
>>> -> for the past couple of years.  It's a python library for structure
>>> -> learning of Bayesian networks from data and prior knowledge.  I've
>>> -> implemented all features for a 1.0 release and have finalized the API.
>>> ->  It has many unique features not found in any other package (python or
>>> -> not), including:
>>> ->
>>> -> * ability to handle interventional data
>>> -> * exact and heuristic methods for handling missing values and hidden
>>> variables
>>> -> * nice AJAX-y html reports
>>> -> * almost-transparent parallel processing on multiple platforms (XGrid,
>>> -> IPython1 and Amazon EC2)
>>> ->     * this is useful when dealing with many hidden variable
>>> ->
>>> -> Pebl includes sphinx-generated documentation and 200+ unittests.  I
>>> -> think it's ready for prime time but would appreciate an informal code
>>> -> review (or just comments) from this group.  Installation is currently
>>> -> a bit painful but I'm working on improving that. Same goes for the
>>> -> quality of the documentation.
>>>
>>> Hi, Abhik,
>>>
>>> the tutorial is cute, but completely uninformative!  What format is the
>>> data in, for example? What does the analysis actually do??  What is the
>>> output?  Can it solve real problems (and what would be an example of
>>> one, and wouldn't that be a great tutorial instead?)
>>>
>>> However, to find out the answer to these questions I would be happy to
>>> have you come give a presentation at MSU, if you're interested.  E-mail
>>> me off-list if so.
>>>
>>> cheers,
>>> --titus
>>> --
>>> C. Titus Brown, ctb at msu.edu
>>>
>
> [This is meant to be critical but not negative]
>
> Please indicate the license in the documentation because I should not have
> to download/install it to find it. One reason is that simply by just
> downloading the package can imply acceptance of the terms. Only by going to
> the project home do I see the license type (MIT) but not the ACTUAL license
> (link is to the general license and does not contain the author or authors
> of PEBL).
>
> How is this different from say OpenBayes (http://www.openbayes.org/) , BNT
> (http://www.cs.ubc.ca/~murphyk/Software/BNT/bnt.html), or, more importantly,
> the SciKit on BNT: http://scipy.org/scipy/scikits/wiki/BayesNet ?
> (Note that your project is linked there.)
> It would be nice to have a single package for this (yes, I can guess what is
> involved).
>
> What is the latest version of PEBL?
> What operating systems are supported and known to work under? If you assume
> Linux or Windows just say so!
> What version of Python is required and supported? Is it 2.5 specific or not
> (NumPy 1.1 and support Python 2.3 or higher; NumPy 1.2 will support Python
> 2.4 or higher)?
> What version of NumPy is required and supported?
>
> What software versions were used in creating the tutorial and installation
> documentation? These should be listed at the start of the tutorial.
>
> Finally, if you want code review, why do you not provide the code?
> Actually I can get via the main project page (but the egg seems to be very
> old) but not a tarball.
>
> Bruce
>
>



-- 
Abhik Shah - http://umich.edu/~shahad
Systems Biology Lab, University of Michigan