[SciPy-Dev] Proposed enhancement: pure Python implementationof LP simplex method (two-phase)

[Pauli Virtanen]

Fri, 30 Jul 2010 18:54:40 +0800, Enzo Michelangeli wrote:
> Thanks for uploading lp2.py: I hadn't noticed it and I uploaded a
> revised lp.py with the changes you required except for the object stuff,
> on which I had a question (namely: should the NamedTuple be based on
> http://docs.python.org/library/collections.html#collections.namedtuple
> ?). Anyway, feel free to remove lp.py .

We need to support also Python 2.4 and 2.5 which do not have 
collections.namedtuple, so we cannot use the pre-cooked one.

    ***

Another thing: you have a magic number 1e-10 in there:

1) Should this tolerance be user-configurable?

2) Should it be a relative tolerance instead of an absolute one?
   I.e., is the quantity H it is compared with always of order 1?


Cheers,
Pauli