[Matrix-SIG] Reading binary data from file into NumPy arrays: numpyio

Paul F. Dubois Paul F. Dubois" <dubois1@llnl.gov
Thu, 18 Jun 1998 09:47:03 -0700

This is part of the dilemma I face with even the parts I have already
inherited in the distribution. With the exception of the core pieces the
rest of the modules such as the lapack, fft, ..., are all just extras. Now
for the random array facility, for example, there is one in the distribution
that looks more "official" than the other one but it isn't of higher
quality, it is just different. Now I see someone posting FFT and binary data
readers; we have an FFT package and we have something that is superior to
binary data for Unix, and it will soon be available on Windows. So how do I
decide what to "bless" and what not to? I suspect blessing and regression
testing all that stuff would soon be a full-time job.

Maybe the right idea is like Amazon.com; you list all the books but have
places for reader and author "reviews". This makes it cavaet emptor but
wasn't it always so? I'm thinking of web site that is a guide to the
contributed packages. Actually, I even think I know how to write such a site
pretty easily. Then we could just have the sources to all the packages
available on an even footing.

This is clearly identical to the general problem of what Guido "blesses" or
not. I think it less likely that you can pick a computer-science area that
Guido doesn't know anything about than you can pick some area of numerical
analysis, probability, and statistics that I know nothing about. Clearly
nobody is capable of knowing all this stuff. The documentation for the NAG
library runs nine volumes.

-----Original Message-----
From: Janko Hauser <jhauser@ifm.uni-kiel.de>
To: hinsen <hinsen@cnrs-orleans.fr>
Cc: matrix-sig@python.org <matrix-sig@python.org>
Date: Thursday, June 18, 1998 9:22 AM
Subject: Re: [Matrix-SIG] Reading binary data from file into NumPy arrays:

>hinsen writes:
> > [Description of a very interesting module deleted.]
> >
> > > If anyone is interested in my source code, let me know and I will send
> > > it to you or let me know where to put it.
> >
> > That again raises the issue of distributing NumPy-related code.
> > How about creating an archive on starship? I'd volunteer to take care
> > of it. Or does someone have a better idea?
>I think we definetly need something like an archive. But I fear that
>we also need some tests for quality (justifing from my own attempts
>to give other some code).
>So my wish is make more code public available for testing and also
>different solutions to a given problem. Than in a second pass collect
>the more useful pieces and make a central archive of them. I fear if
>we only have a contrib-like directory and there is some bad or
>not enough tested code in it, the hole package (NumPy) get bad
>credits. If we can separate this and perhaps also have some ways for
>regression tests for the released code all will look better.
>Matrix-SIG maillist  -  Matrix-SIG@python.org