Beefy Boxes and Bandwidth Generously Provided by pair Networks
go ahead... be a heretic
 
PerlMonks  

Re: Evolving a faster filter?

by roboticus (Canon)
on Jan 05, 2013 at 15:31 UTC ( #1011793=note: print w/ replies, xml ) Need Help??


in reply to Evolving a faster filter?

Ovid:

In the past, I've used a modified priority queue to hold function pointers (code was in C). As the code would run, it would keep the last "successful" function at the first entry in the queue, the rest of the queue sorted by the number of successes so far. So the functions were run in order of their success rate. It worked well, as frequently the last successful function would be successful on the next item in the list. In my case, all the functions had a similar runtime, so I didn't worry about figuring time into the equation. Also, my items to process tended to have similar items grouped together, so one function might be successful several tens of items in a row, so updating the queue wasn't particularly frequent.

Since your functions have significantly different runtimes, I think I'd try a similar procedure, but rather order by success rate, I might try success rate divided by average runtime. If the average runtime is known ahead of time, then the code is pretty straightforward. If your runtime changes, you might try rebalancing the queue every X operations, and use the time spent for X operations to modify the weights of the first few entries.

Finally, have you tried swapping your filters vs objects loops? If you have few filters and many objects, I'd guess that it might be faster to do it like:

sub remove_unwanted { my ( $self, @objects ) = @; my @rv; OUTER: for my $curObj (@objects) { for my $predicate ($self->predicates) { next OUTER unless $self->$predicate($curObj); } push @rv, $curObj; } return @rv; }

Of course, you'd have to rearrange a good deal of code if you were going to try it this way. I recognize your handle, so I suspect you've already evaluated the tradeoff between filtering a list of objects vs. evaluating single objects. I'm mostly mentioning this trick for others with similar problems to solve.

...roboticus

When your only tool is a hammer, all problems look like your thumb.


Comment on Re: Evolving a faster filter?
Download Code

Log In?
Username:
Password:

What's my password?
Create A New User
Node Status?
node history
Node Type: note [id://1011793]
help
Chatterbox?
and the web crawler heard nothing...

How do I use this? | Other CB clients
Other Users?
Others exploiting the Monastery: (6)
As of 2014-10-25 03:03 GMT
Sections?
Information?
Find Nodes?
Leftovers?
    Voting Booth?

    For retirement, I am banking on:










    Results (141 votes), past polls