[SciPy-Dev] LinearOperator and new solver

[Ralf Gommers]

On Wed, Dec 12, 2018 at 11:24 AM Matteo Ravasi <matteoravasi at gmail.com>
wrote:

> Dear Scipy-dev community,
>
> I would like to bring forward two proposals related to your
> scipy.sparse.linalg.LinearOperator class and linear solvers in general:
>
>
>
>    - LinearOperator: I am the main developer and maintainer of the PyLops
>    library that has been recently open-sourced (git repo:
>    https://github.com/Statoil/pylops, doc:
>    https://pylops.readthedocs.io/en/latest/index.html)
>    <https://pylops.readthedocs.io/en/latest/index.html>. As you will see
>    I heavily rely on your LinearOperator class and build on top of it creating
>    various basic linear operators and more specific ones for signal processing
>    and geoscience applications of inverse problems. When I started working on
>    this I was surprised to find very little information and examples online on
>    how to use your LinearOperator and the possibility to subclass it was only
>    mentioned in one line of your documentation and nowhere else online to my
>    knowledge. I wonder if you would consider pointing to PyLops in your
>    documentation to facilitate users i) to know how to get started with
>    LinearOperator, ii) avoid rebuilding the wheel if what they are after can
>    be done or is done already in PyLops. I find your class fantastic and saved
>    me lots of time but I feel is one of those things that few people realize
>    exist and understand how to use it ;)
>
> That sounds like a good idea to me. A link from the See Also part of the
LinearOperator seems appropriate to me. I don't see a good place in the
tutorial section of our docs to put a link, because LinearOperator isn't
really mentioned there. If you'd like to write a short section in, e.g.,
the ARPACK tutorial and add a link there too, that would be great.


>
>    - Solvers: I would like to know if you would be interested to add a
>    new solver to your suite of linear solvers. Specifically the solver is
>    called SPGL1 and it is a very popular solver in the mathematical community
>    for sparsity-promoting linear optimization. It was developed at UBC
>    (University of British Columbia) https://www.cs.ubc.ca/~mpf/spgl1/ and
>    has a Matlab open-source code.
>
> That code is GPL licensed, so we cannot accept any code derived from it.
If the original author wouldn't mind relicensing or giving explicit
permission for the Python port to be BSD or MIT licensed, then we can
consider it.

Cheers,
Ralf


>
>    - I have been thinking about porting it to python for a while and
>    recently another python user started a git repo called
>    https://github.com/drrelyea/SPGL1_python_port. However this is
>    something I would more naturally see as part of scipy instead of an
>    indipendent library. I am willing to contact the author of this repo and
>    help him out directly to finish the porting and make it to (or close to)
>    scipy standards... so far the code and repo is not ready to be included in
>    professional code like scipy in my opinion. But I would like to hear what
>    you think, if you would consider adding this to scipy once in good shape or
>    if you think this is not in the scope of scipy library :)
>
>
> Looking forward to hearing from you,
>
> Best wishes
>
> MR
> _______________________________________________
> SciPy-Dev mailing list
> SciPy-Dev at python.org
> https://mail.python.org/mailman/listinfo/scipy-dev
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.python.org/pipermail/scipy-dev/attachments/20181212/99595bb5/attachment.html>