nemesis wrote:
> Mueen Nawaz escreveu:
>>     Python's gaining ground with a lot of social scientists in
>> academia. I
>> suspect it's ease of use without hindering the programmer (other than
>> via performance) is the reason.
> 
> There's scypy and numpy for people who want good performance and
> higher-level semantics. :)

	Yeah, but not on par with C, etc.

-- 
"Now we all know map companies hire guys who specialize in making map
folding a physical impossibility" - Adult Kevin Arnold in "Wonder Years"


                    /\  /\               /\  /
                   /  \/  \ u e e n     /  \/  a w a z
                       >>>>>>mue### [at] nawazorg<<<<<<
                                   anl

Post a reply to this message

Mueen Nawaz escreveu:
> nemesis wrote:
>> Mueen Nawaz escreveu:
>>>     Python's gaining ground with a lot of social scientists in
>>> academia. I
>>> suspect it's ease of use without hindering the programmer (other than
>>> via performance) is the reason.
>> There's scypy and numpy for people who want good performance and
>> higher-level semantics. :)
> 
> 	Yeah, but not on par with C, etc.

They are just thin wrappers for scientific/numeric libs written in C.

-- 
a game sig: http://tinyurl.com/d3rxz9

Post a reply to this message

nemesis wrote:
> They are just thin wrappers for scientific/numeric libs written in C.

	Well, you may get comparable performance if you never do, say, a loop.
Or if you can manipulate large datasets using only the vectorized forms
and not make use of almost anything within the actual Python language.
In real world cases, it's unlikely.

	It's the same with MATLAB - they're just wrappers for libraries in
C/C++. Yet few perceive it to be as fast. You can always concoct
examples where they're comparable, but in reality, you may not be able
to vectorize everything.

	I often tell people that NumPy/SciPy is a "competitor" of MATLAB (and
has a much faster interpreter, which can make a big difference). If
MATLAB doesn't perform well for them, and if they properly know how to
use MATLAB, then I don't recommend NumPy/SciPy to them either.

-- 
"Now we all know map companies hire guys who specialize in making map
folding a physical impossibility" - Adult Kevin Arnold in "Wonder Years"


                    /\  /\               /\  /
                   /  \/  \ u e e n     /  \/  a w a z
                       >>>>>>mue### [at] nawazorg<<<<<<
                                   anl

Post a reply to this message

Mueen Nawaz <m.n### [at] ieeeorg> wrote:
> nemesis wrote:
> > There's scypy and numpy for people who want good performance and
> > higher-level semantics. :)
>
>  Yeah, but not on par with C, etc.

Vectorize, vectorize, vectorize!

 - Ricky

Post a reply to this message

triple_r wrote:
> Vectorize, vectorize, vectorize!

	See my other post.

-- 
"Now we all know map companies hire guys who specialize in making map
folding a physical impossibility" - Adult Kevin Arnold in "Wonder Years"


                    /\  /\               /\  /
                   /  \/  \ u e e n     /  \/  a w a z
                       >>>>>>mue### [at] nawazorg<<<<<<
                                   anl

Post a reply to this message

triple_r wrote:
> Mueen Nawaz <m.n### [at] ieeeorg> wrote:
>> nemesis wrote:
>>> There's scypy and numpy for people who want good performance and
>>> higher-level semantics. :)
>>  Yeah, but not on par with C, etc.
> 
> Vectorize, vectorize, vectorize!

	I hate to argue like this, but let me ask an irritating question. Why
do a lot of people use C/C++ for heavy computation if they could do it
with MATLAB (cost is not always an issue, and a number of these people
aren't necessarily dealing with huge datasets)?

	I doubt the reason for most of them is that they don't know about
vectorize.

-- 
"Now we all know map companies hire guys who specialize in making map
folding a physical impossibility" - Adult Kevin Arnold in "Wonder Years"


                    /\  /\               /\  /
                   /  \/  \ u e e n     /  \/  a w a z
                       >>>>>>mue### [at] nawazorg<<<<<<
                                   anl

Post a reply to this message

Mueen Nawaz <m.n### [at] ieeeorg> wrote:
> triple_r wrote:
> > Vectorize, vectorize, vectorize!
>
>  See my other post.

Sorry.  I read from top to bottom and commented as I went, but I do see that
this is a bit redundant.  Repetitive, too.

 - Ricky

Post a reply to this message

Mueen Nawaz <m.n### [at] ieeeorg> wrote:
> triple_r wrote:
>  I hate to argue like this, but let me ask an irritating question.

I think we're mostly in agreement about this.  I find that most of the data
processing I do, including conditionals and such, can be vectorized just fine,
although you're correct in saying that some things just don't vectorize.

> Why
> do a lot of people use C/C++ for heavy computation if they could do it
> with MATLAB

I think there's comfort in doing the low-level details yourself.  I'm guilty of
this, and only just figuring out that most of what I spend my time doing has
already been done pretty well in a high-level language.

>  I doubt the reason for most of them is that they don't know about
> vectorize.

True, although in my experience, most people don't.  The first language I ever
learned was actually the POV-Ray SDL, and it really stuck with me to do all of
the low-level details yourself, relying only on some basic functions.  It works
pretty well for POV-Ray scenes since most don't involve heavy processing, but
sooner or later you start to run into problems.  It's the classic, "If all you
have is a hammer..." scenario.

 - Ricky

Post a reply to this message

Making a language easy is not really an issue, if it's too low level, someone
will aether build a library or a simpler language on top of it.

If you want something simpler, make your compiler that generates DSL. Just like
a M$ Workflow diagram "gets compiled to" (generates) C#, which gets compiled
into IL which at last gets compiled into a CPU instruction set.

If you don't like keyboards, code with your mouse using a modeler :-)

Post a reply to this message

Louis schrieb:
> Making a language easy is not really an issue, if it's too low level, someone
> will aether build a library or a simpler language on top of it.
> 
> If you want something simpler, make your compiler that generates DSL. Just like
> a M$ Workflow diagram "gets compiled to" (generates) C#, which gets compiled
> into IL which at last gets compiled into a CPU instruction set.
> 
> If you don't like keyboards, code with your mouse using a modeler :-)

It would be of quite some benefit to be able to share scene files and 
include files between POV-Ray users, without needing to get half a dozen 
or more different libraries, on-top languages and what-have-you-nots to 
cooperate.

(Even if your suggested approach would be the way to go, it would not 
change the fact that thought should be put into the design of the 
language - if not how to make it easiest for users, then how to make it 
most flexible for adding libraries or auto-generating code.)

Post a reply to this message