David - On Wed, Jul 27, 2011 at 8:21 PM, David Myers <[email protected]> wrote: > Dave, > > I think BrowserUK's point was functional efficiency as opposed to > operators. The matrices in question had values that were zeroes and ones > only, probably connectivity matrices from some kind of graph. As such > they can be implemented within regular Perl via bit-vectors, and such an > implementation is both space efficient and fast. > > PDL, and I may be wrong here, is optimized for floating point > efficiency. I'd be happy to be proven wrong in this, but I think, if you > get past all the extraneous questions about operators in PDL, that the > compact expression of 0 & 1 only matrices as bit-vectors was at the > heart of UK's comments. > > That you could count bit wise operations with something like this one > liner > > my $count = sum ($poweroftwo & $My_PDL_Matrix ) >> $power; > > Is yet another issue. > > Sincerely, > > David Myers (dwm042@perlmonks). > CPAN ID DWMYERS
Yes, that was his point, but since he raised the question, it needed to be answered. As for whether pure Perl would beat PDL on this one, though I will be working with networks and adjacency matrices starting this fall, I haven't started yet and I don't have time to cook up examples and run benchmarks. It might be a toss-up: it may depend on the size of the adjacency matrix. David _______________________________________________ Perldl mailing list [email protected] http://mailman.jach.hawaii.edu/mailman/listinfo/perldl
