On Sun, May 31, 2009 at 7:31 PM, Chris Marshall <[email protected]> wrote: > Gabor Szabo wrote: >> ... >> >> Certainly price is a big difference and the >> fact that PDL is open source while Matlab is >> proprietary so I am more interested in the >> technical aspects. > > I use PDL and Matlab at work. At home I can > only afford to run PDL.
Good point but I doubt any of the people I know would want to use it at home. This is work for them and most of them are not the hacker type. > >> I searched a bit and found this article: >> >> http://www.freesoftwaremagazine.com/articles/cool_fractals_with_perl_pdl_a_benchmark >> which on the last page shows that PDL is twice as >> fast as Matlab - at least in that specific >> Mandelbrot example. > > Benchmarks are a tricky business. While PDL came > out well here, a different problem measured might > result in a different outcome. The key point here > is that programming at a higher level of abstraction > can give performance and ease of coding. For many > problems, the software development and algorithm > development and coding times far outweigh the runtime. Yeah, I know there are usually all kinds of Benchmarks showing sometimes this sometimes that faster but wait, do you say that it is faster to write in Perl than in Matlab ? I know it is much faster to write in Perl than in C or Java but I know nothing about Matlab. I know Perl has the vast library on CPAN and Matlab, even if it has third party libraries they are probably proprietary too but are those modules relevant for the usual tasks done by Matlab people ? Can you point out areas where Matlab is used but it clearly not strong enough? (e.g. someone told me they use Perl to pre-process large log-files to be in I think csv format as parsing that large file would be prohibitively slow in Matlab) >> I found a few hits mentioning PDL::Matlab but >> no actual code. > > Right. This is planned to be a set of routines to > allow more matlab like syntax and functions for PDL. > This would make it easier for Matlab users to start > with PDL and easier for PDL users to go back and > forth between PDL and Matlab---as I often do. > > No code yet. > >> http://en.wikipedia.org/wiki/Perl_Data_Language >> says "[PDL is ] intended for small to >> medium sized analysis tasks" > > Pretty vague there since one person's small > analysis task could be another person's huge > task. Yes, you are right. Maybe some harder data points would be worth adding on that page or people will use it as a fact. > >> I could not find other references. >> >> So I am really interested you people pointing >> me to further references or giving your opinion. > > An advantage that Matlab has is lots of > pre-rolled algorithms that users can work with > right away, without even knowing what or how > they work. Here we go. So if I talk to these Matlab using people and mention them they could switch to PDL what can I expect? What will they ask and how am I going to answer? They will ask all kinds of questions like Does PDL support X ? so what are those X-es people might immediately seek and how do Matlab and PDL compare on those? > > Because PDL is Perl based, it makes it easy > to add additional functionality from external > libraries. The threading capabilities of PDL > also make it possible to implement processing > across multiple dimensions easily with good > performance. > > The catch is that extending PDL is more than > "beginner's work". > > To that end, we've been working to improve > the portability and robustness of PDL as a > distribution to allow more folks (not just > research scientists or Perl programmers) to > just start out immediately with PDL. The > current PDL-2.4.4 is the most portable PDL > ever. It builds on win32, darwin, liunx, > solaris, *bsd,... Gabor _______________________________________________ Perldl mailing list [email protected] http://mailman.jach.hawaii.edu/mailman/listinfo/perldl
