Octave review

[Jordi Gutiérrez Hermoso]

2009/2/7  <bharat at arithos.com>:
> More feathers will be added to the review if some benchmark
> data is available.

Benchmarking is nice, but be careful to do it correctly. In particular,

> Like speed of execution,

This one is fairly objective, but make sure to benchmark things that
are actually commonly used. E.g. a FFT, a matrix exponential, a
particular piece of identical code with many for-loops...

> ease of use, learning curve etc.

These are very subjective and not subject to benchmarking unless you
run some sort of sociological and psychological experiment that is
almost impossible to perform, since you need specialised test subjects
that are very hard to find (i.e. people with a solid abstract
foundational knowledge of numerical analysis but who have never used
any of the options you want to benchmark).

> I am planning to benchmark Matlab, Octave,
> Scilab and Python

Octave and Matlab are easy to benchmark because they can run the exact
same code for many examples, but the others have different syntax, and
depending on how you translate the Octave and Matlab code into Scilab
and Python, you may get different performance, just as you would get
different performance within Octave itself if you were to vectorise
the code.

- Jordi G. H.