[SciPy-User] Least-squares fittings with bounds: why is scipy not up to the task?

[Gael Varoquaux]

David,

You raise good points. On the one hand, contributing to scipy may be a
bit more technical than it should be for someone wanting to add simple
things: it requires building the beast! On the other hand, a lot of the
points you raise simply boil down to the fact that putting independent
piece of code together in a larger, somewhat consistent project, is
actually much more work than writing the individual pieces of code. There
is a large overhead of big project. It is also much more value.

With regards to the scikits, that you mention, I am a huge fan of the
scikits approach, because it enables to break down a bit that friction of
large project, to the cost of some fragmentation. Let us not fool us, the
major scikits can easily grow fat and end up in the same situation,
although if I believe the N^2 complexity increase (
http://www.computer.org/csdl/trans/ts/1979/02/01702600-abs.html ) means
that the union of two projects will tend to be significantly harder to
handle that each project separately.

To answer your question about scikits: they are nothing much more than a
brand name, and maybe a bit of a community trend. Both are good, because
they help create dynamism, but they are not enough per se.

With regards to how disorganized things are, you are definitely right.
Organizing a community, writing contribution guidelines, keeping web
pages up to date, takes a lot of time. Such work, also from volunteers,
is also needed to keep a project alive, and even more a 'meta project',
like scipy.

Gaël

PS: As a side note, I am recruiting on a regular basis (say one young
engineer everything year) people to work part time on the scipy ecosystem
(mainly on scikit-learn, but side projects are encouraged). The salary
doesn't compete with what the industry has to offer, and I've had a hard
time finding good people.