Re: PSPP

[Jason Stover]

Hi Mythreyi,

Thanks for the interest. PSPP could use more development
of its statistical procedures.

I have written a basic least-squares fitting library and a rudimentary
regression procedure that uses it. It does not produce much output
yet. Before I check the code in to savannah, I need to clean it
up. You can see what I've written at www.sakla.net/pspp_contrib.tar.gz
and www.sakla.net/linreg.tar.gz. Ben sent email detailing the
necessary clean-up. Check the 'contribution and cvs' thread in the
mail list archive at:

http://lists.gnu.org/archive/html/pspp-dev/2005-08/msg00051.html

That said, how you contribute is mostly up to you. The GNU
volunteer coordinator told me that the best project to which to
contribute is one that you use, so I would begin by anonymously
checking out pspp at savannah, then using it for your own purpose. If
you need a linear regression procedure first, you can test or clean up or add
to the code I already wrote. If you need a different, unimplemented
procedure, investigate how to implement it. (I recommend not adding
much to the code I wrote until it's in savannah. That means I had
better get busy fixing what I wrote.)

As you mentioned, PSPP lacks modeling procedures, including linear
regression, generalized linear models, splines, etc. There are plenty
of such procedures to write. Before getting started, take a look at
the linreg library I wrote, especially the pspp_linreg_cache struct. I
hope all PSPP's modeling procedures will be as modular as linreg, so
that in the guts of PSPP a 'model' will be a structure usable by
different procedures, including, e.g., a boosting procedure that
allows users to tie several models together. This kind of design would
allow PSPP to be an extensible 'data mining' tool in addition to being a
clone of SPSS. 

PSPP needs more documentation. It has a nice syntax manual, but lacks
documents which explain the statistics behind the procedures. I think
Ben wants to include such documentation.

If you want to write code, you should read the GNU coding standards,
if you haven't done so already. 

To contribute any more than small patches, FSF requires all
contributors and their employers to sign some legal papers assigning
copyrights to FSF.

Let us know if you need anything else. 

-Jason

On Mon, Aug 29, 2005 at 08:10:58PM -0700, Mythreyi Bhargavan wrote:
> Hi Jason,
> 
> I'd like to help out with the PSPP regression
> procedures.  I can interpret statistical procedures,
> or figure out algorithms, and maybe even write some
> code.  I was told that you have been writing the
> regression procedures and are fairly far along with
> them.  So I thought I'd coordinate with you and figure
> out how I could be most useful.
> 
> Mythreyi
> 
> __________________________________________________
> Do You Yahoo!?
> Tired of spam?  Yahoo! Mail has the best spam protection around 
> http://mail.yahoo.com 

-- 
address@hidden
SDF Public Access UNIX System - http://sdf.lonestar.org