[SciPy-Dev] scikits datasmooth (was: scikits contribution?)

[Jonathan Stickel]

On 7/16/10 11:55 , josef.pktd at gmail.com wrote:
>>> It would be better to upload also the sdist,
>>> >>  Since scikits.datasmooth is pure python, an egg doesn't really have an
>>> >>  advantage and is specific to a python version, py2.6 only.
>>> >>  (I'm on py25)
>>> >>
>>> >>  Why do you have the code duplication in two modules and the
>>> >>  conditional import in __init__.py ?
>>> >>  try:
>>> >>       from regularsmooth import *
>>> >>  except ImportError as error:
>>> >>
>>> >>  What is the import error that might occur with regularsmooth ?
>>> >>
>>> >>  Also an example to quickly try out the scikit would be very useful.
>>> >>  (And tests and docs are not included yet.)
>>> >>
>> >
>> >  Thanks for the feedback.  I was confused what was the "sdist", but I think I
>> >  figured it out and uploaded it.
>> >
>> >  I will get to examples, tests, and docs later when I have time (this will
>> >  require more learning for me).  The functions are well documented and the
>> >  primary use function "smooth_data" shows an example.
>> >
>> >  There are two primary implementations: "smooth_data" that includes
>> >  cross-validation, and "smooth_data_constr" which will take constraints but
>> >  does not include cross-validation.  The later requires the module "cvxopt".
>> >    If this is not available, I wanted to allow a user to still use the
>> >  unconstrained smoother.  Maybe there is a better way to do this than my
>> >  "try/except" hack.  Suggestions are welcome.
> Try/except are fine for this, but it wasn't very informative. You
> could also try/except import cvxopt, or add a note that the full
> version requires cvxopt. Maybe it's noted somewhere in your code, I
> haven't looked that carefully.

Kind of noted in the README.  I am thinking of putting the try/except in 
the script itself rather than in __init__.py, but I will get to it later.

>
> Is it possible to replace cvxopt with scipy optimizers? But this is
> not necessary for a scikit.
>

I needed a quadratic programming (qp) implementation, and I couldn't 
find any qp in scipy.  I found the one in cvxopt to be quite good.

Jonathan