Le 4 avril 2012 07:31, Mathieu Blondel <[email protected]> a écrit :
> Creating a new language has its advantages (e.g., tailor the language to
> scientific computing) but the main disadvantage is that they need to
> recreate a complete ecosystem (plotting, interactive shell, ...). I wonder
> if we could get most of the advantages of julia in Python with Numba
> (https://github.com/ContinuumIO/numba) or PyPy.

That ability to do lisp-style macro manipulating the AST can be very
interesting for runtime-dependent code generation / optimization (e.g.
leveraging GPU when there, switch between cluster distribution and
multiple cores depending on both the infrastructure and the dataset
size...). This can be hacked in python but this will never feel as
natural as in lisp (or using compiler extensions as in scala). And I
agree that it's probably a good idea to keep python as simple as it is
in that respect (no macros / runtime AST modification / DSL support).
It can make julia very powerful in some situations though.

I don't like the default namespace pollution with domain specific
functions and the lack of a clean module system. IMHO, explicit
imports of symbols in a namespace is very important for
maintainability once you pass the "single experimental script" step in
a project and readability counts, explicit is better than implicit...

-- 
Olivier
http://twitter.com/ogrisel - http://github.com/ogrisel

------------------------------------------------------------------------------
Better than sec? Nothing is better than sec when it comes to
monitoring Big Data applications. Try Boundary one-second 
resolution app monitoring today. Free.
http://p.sf.net/sfu/Boundary-dev2dev
_______________________________________________
Scikit-learn-general mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/scikit-learn-general

Reply via email to