Re^7: Longest Increasing Subset

wow... so I implemented the wiki page's pseudocode, and benchmarked my code vs the Longest Increasing Subsequence:

results:

20 [953 317 563 582 789 629 613 918 9 680 276 163 119 915 967 426 820 
+227 256 650]
lis: [ 317 563 582 613 680 915 967 ]
pryrt: [ 317 563 582 629 680 915 967 ]
Benchmark:
running
  lis, pryrt
 for at least 10 CPU seconds
...

       lis: 10 wallclock secs (10.51 usr +  0.00 sys = 10.51 CPU) @ 25
+210.27/s (n=265086)

       pryrt: 11 wallclock secs (11.26 usr +  0.00 sys = 11.26 CPU) @ 
+ 0.18/s (n=2)

            (warning: too few iterations for a reliable count)

         s/iter       pryrt     lis
pryrt      5.63        --     -100%
lis   3.97e-005 14197064%        --
[download]

With a random list just twenty elements long, it went from 5-6s per run for my code, to better than 25_000 runs per second with the efficient algorithm: that's almost 150_000 times faster, and that's just with 20 elements in the array. (My original benchmark tried cmpthese() on random length arrays, using the defaults to my genarray() function, but after a minute, I didn't feel like waiting that long, so switched to a 10sec timethese()/cmpthese() pair. Multiple runs give similar results.)

When run-time is money, it pays to search for known-good algorithms, rather than just trusting that you've thought of an efficient algorithm!

Thanks ++gilthoniel for an interesting thread, and ++BrowserUK, for once again bringing his knowledge of efficient coding practices to help us all.

To all who come upon this thread: definitely use the efficient code, rather than my original code!

Comment on Re^7: Longest Increasing Subset Select or Download Code

Replies are listed 'Best First'.
Re^8: Longest Increasing Subset by BrowserUk (Patriarch) on Oct 24, 2016 at 16:06 UTC
I also implemented the wiki pseudo-code, resulting almost identical code: <Reveal this spoiler or all in this thread> Though I pass the data in via a reference for efficiency. I've been trying to wrap my brain around the mentioned Knuth optimisation without success. One comment on your code. This `return wantarray ? @S : [@S];` throws away the advantage of returning a reference, by constructing a list from the existing array, which is then used to construct another (anonymous) array, a reference to which is returned. Better to just do `return wantarray ? @S : \@S;` With the rise and rise of 'Social' network sites: 'Computers are making people easier to use everyday' Examine what is said, not who speaks -- Silence betokens consent -- Love the truth but pardon error. "Science is about questioning the status quo. Questioning authority". I knew I was on the right track :) In the absence of evidence, opinion is indistinguishable from prejudice.	[reply] [d/l] [select]

Replies are listed 'Best First'.

Re^8: Longest Increasing Subset
by BrowserUk (Patriarch) on Oct 24, 2016 at 16:06 UTC

I also implemented the wiki pseudo-code, resulting almost identical code:

Though I pass the data in via a reference for efficiency.

I've been trying to wrap my brain around the mentioned Knuth optimisation without success.

One comment on your code. This return wantarray ? @S : [@S]; throws away the advantage of returning a reference, by constructing a list from the existing array, which is then used to construct another (anonymous) array, a reference to which is returned.

Better to just do return wantarray ? @S : \@S;

With the rise and rise of 'Social' network sites: 'Computers are making people easier to use everyday'

Examine what is said, not who speaks -- Silence betokens consent -- Love the truth but pardon error.

"Science is about questioning the status quo. Questioning authority". I knew I was on the right track :)

In the absence of evidence, opinion is indistinguishable from prejudice.

[reply]
[d/l]
[select]