Re^2: When the Best Solution Isn't

sub qshuf {
    sort { .5 <=> rand(1) } @_;
}

cmpthese(10, {
    slice    => sub { slice @array },
    xform    => sub { xform @array },
    shufl    => sub { shufl @array },
    qshuf    => sub { qshuf @array },
});

        Rate  xform  slice  shufl  qshuf
xform 14.4/s     --   -38%   -70%  -100%
slice 23.3/s    61%     --   -51%   -99%
shufl 47.8/s   232%   105%     --   -99%
qshuf 3284/s 22681% 14008%  6767%     --
[download]

Mindboggling. "qshuf", an N log N algorithm, shows to be 67 faster than any of linear algorithms, and you don't raise a flag?

Your benchmark is extremely flawed. You're running your sort in void context. Perl, being smart, knows that sort in void context isn't going to do anything useful, so it's not going to work up a sweat, and just won't do anything.

Benchmark: timing 10000 iterations of qshuf, shufl, slice, xform

Hmmm, with a first argument of '10' to 'cmpthese'? Either your output doesn't match your program, or your Benchmark.pm module is different than mine.

Comment on Re^2: When the Best Solution Isn't Download Code

Replies are listed 'Best First'.
Re^3: When the Best Solution Isn't by BrowserUk (Patriarch) on Apr 01, 2005 at 10:28 UTC
Mindboggling. "qshuf", an N log N algorithm, ... Hmm. N log N with N=4 == 2.408. How does that affect your boogle? The main part of the post was the statistical analysis, but you are correct, the void context does make a difference to the timing and the output posted was obviously an accumulation from 2 different runs of the program which had been modified in the interim. Hmmm, with a first argument of '10' to 'cmpthese'? Either your output doesn't match your program It was a long time ago, so I am guessing more than remembering, but I think what probably happened was that I wrote the benchmark code first, ran it (with the iters counter set to 10000) and c&p'd the output to the source file. Seeing that qshuf() seemed very fast, (I probably wasn't even aware that sort silently optimised for void context rather than doing something sensible like issuing the "Useless in a void context" warning as other parts of perl do). I then went on to add the statistical analysis, and dropped the iterations counter to 10 to save time while getting the formatting of the output right. Here is a an updated version (run on 5.6.1 as the original was): Read more... (5 kB) And by way of recompense, here's an updated version incorporating List::Util::shuffle run on 5.8 Read more... (5 kB) Thanks for pointing out the discrepancies--even if it did take 2 years for collected populace to notice :) Examine what is said, not who speaks -- Silence betokens consent -- Love the truth but pardon error. Lingua non convalesco, consenesco et abolesco. Rule 1 has a caveat! -- Who broke the cabal?	[reply] [d/l] [select]
Re^4: When the Best Solution Isn't by Anonymous Monk on Apr 01, 2005 at 11:58 UTC
Hmm. N log N with N=4 == 2.408. How does that affect your boogle? Not at all. Three points: In CS, when people talk about `log`, they mean `²log`, and `4 ²log 4 == 8`. I don't know what significance 4 has. The array has a 1000 elements, so N equals 1000, so `N log N` would be about 9966. But numbers don't mean anything. When we talk about a linear, or a quadratic, or a N log N, or some other function of N algorithm, we talk about how it scales. Any constants involved are ignored.	[reply]
Re^5: When the Best Solution Isn't by BrowserUk (Patriarch) on Apr 01, 2005 at 12:02 UTC
You didn't notice the smiley? I conceded, and thanked you for, all your points. And posted updated code and stats. Examine what is said, not who speaks -- Silence betokens consent -- Love the truth but pardon error. Lingua non convalesco, consenesco et abolesco. Rule 1 has a caveat! -- Who broke the cabal?	[reply]


Pathologically Eclectic Rubbish Lister
	PerlMonks