Re: [sage-devel] MATLAB: viable alternative...?

[Jordi Gutiérrez Hermoso]

I'm cross-posting this to the Octave maintainers' mailing list in case
other maintainers also want to comment on it or correct anything I say
here.

On 15 August 2011 14:58, William Stein <address@hidden> wrote:

> If somebody walked up to *you* and asked: "Is Sage now a viable
> alternative to MATLAB?" what would you say? I'm especially
> interested in what people who do numerical/applied computation
> think.

With Octave we deal with people like this all the time. They basically
want Matlab, not a substitute, because their
instructor/employer/whoever has told them it must be Matlab and it
can't be anything else. They often also have code written in Matlab
that can't be translated. In the case of running the exact same code,
which happens very frequently with Matlab, I can tell you that for
these people rewriting their code in Scipy isn't an alternative (I am
obviously also referrring to Numpy, as the core component of Scipy).

One of the most frequent requests we get in Octave and then everyone
will magically jump ship from Matlab is to implement a GUI (or an
editor, or a profiler, or an integrated debugger, or all of Simulink).
They want it to look almost exactly like Matlab too, or else they
won't use it. For a long time we've ignored this request, but it comes
up so often that it's hard to keep ignoring it, and a
Matlabish-looking GUI is a release goal for our next version, and the
work is well underway.

There are also a lot of people for whom Matlab is only incidental and
what they really want is Simulink. For this, neither scipy nor Octave
have any sort of alternative, but I understand Scilab has something
called Xcos that does sort of work like Simulink.

Matlab users also tend to have highly specific requests. They want the
*exact* same function from some particular Matlab toolbox and they
want it to work the same way in Octave (or presumably, a direct
equivalent in Scipy), even for silly things like rref because, again,
their instructor/employer/whoever told them that that is the function
they must use. I have recently started working with Scipy, and my
impression is that it has smaller set of contributed functions than
Octave and smaller than Matlab, so when someone from Matlab wants this
or that low-pass filter and accepts no substitute, they get easily
frustrated with the free alternatives and go back to Matlab. A big gap
in Scipy due to silly licencing issues is FFTW, which is the de-facto
standard FFT library both Octave and Matlab use; FFTPACK is an ok
substitute and all that Scipy has. Signal processing is one of
Matlab's biggest use cases and a poor FFT implementation is
problematic.

I imagine demanding users like this exist for the other three big M's
too, but Matlab to me seems to have a well-entrenched niche separate
from the other three M's whoh at least somewhat compete with each
other, whereas Matlab has no direct big competitor. The competition in
the other three M's has at least opened up the users of those CASes to
the possibility of learning a new syntax, but my impression is Matlab
users are less flexible in this regard.

HTH,
- Jordi G. H.
  GNU Octave developer