[Haskell-cafe] Help me speed up my program... or back to the
issue of memoization
Sterling Clover
s.clover at gmail.com
Mon May 5 19:21:22 EDT 2008
I actually suspect that your "supremum" and "infimum" functions are the
problem here -- they look like they might be accumulating thunks and blowing
your stack. But beyond this, they're also o(n) for what could be effectively
an o(log n) operation at the least, if you used ordered sets.
I'd start improving performance here with some profiling, and then some
strictness annotations, and then go from there.
See http://www.haskell.org/haskellwiki/Performance for a whole bunch of good
tips.
--Sterl
On Mon, May 5, 2008 at 11:57 AM, Kirill Kuvaldin <kirill.kuvaldin at gmail.com>
wrote:
> Hello,
>
> I wrote a program in haskell that works with lattice finite automata
> (the generalization of notion of finite state automata). Let me post
> the source code as it is not that long...
>
> The problem is that my algorithm that computes the run (see function
> fun) of an automaton on the given word is not very optimal and takes a
> loooong time as the input word gets larger... (e.g. try this one "run
> m3 "11101101110001010101" ")
>
> Due to the nature of every haskell function being a referentially
> transparent, I think I could have speeded up
> its performance using memoization.
>
> I've read the page on haskell wiki
> (http://www.haskell.org/haskellwiki/Memoization) but it didn't help me
> because it looks I have to modify the actual function source code to
> make use of memoized values.
> What I'm looking for is a kind of a general solution (say, silver
> bullet :) ) so that I will be able to use my function like
>
> > new_run = memoize run
>
> and the results of the "new_run" get automatically memoized. Probably
> it makes sense to memoize deltaext func as well.
>
> Is that possible to do that in Haskell??
>
> Thanks a lot!
> Kirill
>
>
> ======= SOURCE CODE =====
>
> -- data type for lattice
> data Lattice l = Lattice
> [l] -- set of lattice elements
> (l -> l -> l) -- supremum operation
> (l -> l -> l) -- infimum operation
>
> -- returns the lowest lattice element
> lattice0 (Lattice l s i) = l !! 0
> -- returns the greatest lattice element
> lattice1 (Lattice l s i) = l !! ((length l)-1)
>
> -- supremum of 2 lattice elements
> sup (Lattice l s i) x y = s x y
> -- infimum of 2 lattice elements
> inf (Lattice l s i) x y = i x y
>
>
> supremum (Lattice l sup inf) [] = lattice0 (Lattice l sup inf)
> supremum (Lattice l sup inf) (x:xs) = sup x (supremum (Lattice l sup inf)
> xs)
>
> infimum (Lattice l sup inf) [] = lattice1 (Lattice l sup inf)
> infimum (Lattice l sup inf) (x:xs) = inf x (infimum (Lattice l sup inf)
> xs)
> inf3 (Lattice l s i) x y z = infimum (Lattice l s i) [x,y,z]
>
> --- data type for Lattice Automata (LA)
> data LA l state sym = LA
> (Lattice l) -- lattice
> [state] -- set of states
> [sym] -- alphabet
> (state -> sym -> state -> l) -- fuzzy transition function
> (state -> l) -- fuzzy initial state
> (state -> l) -- fuzzy final state
>
> --- extended transition function
> deltaext :: (Eq state) => (LA l state sym) -> state -> [sym] -> state -> l
> deltaext (LA l states chars delta sigma0 sigma1) x [] y =
> if x == y then (lattice1 l) else (lattice0 l)
> deltaext la@(LA l states chars delta sigma0 sigma1) x (a:w) y =
> supremum l
> [ inf l
> (delta x a z)
> (deltaext la z w y)
> | z <- states]
>
> -- runs the Lattice Automaton on the given word
> run la@(LA l states chars delta sigma0 sigma1) w =
> supremum l
> [ inf3 l
> (sigma0 x)
> (deltaext la x w y)
> (sigma1 y) | x <- states, y <- states]
>
> ---
> --- examples
> ---
>
> l3 = Lattice [0.0, 0.5, 1.0] max min where
> max x y = if x > y then x else y
> min x y = if x < y then x else y
>
> m3 = LA l3 ['a', 'b'] ['0', '1'] delta sigma0 sigma1 where
> delta 'a' '0' 'a' = 1
> delta 'a' '0' 'b' = 0.5
> delta 'b' '0' 'a' = 0.5
> delta 'b' '0' 'b' = 1
> delta 'a' '1' 'a' = 0
> delta 'a' '1' 'b' = 1
> delta 'b' '1' 'a' = 1
> delta 'b' '1' 'b' = 1
> sigma0 'a' = 1
> sigma0 'b' = 0.5
> sigma1 'a' = 0.5
> sigma1 'b' = 1
> _______________________________________________
> Haskell-Cafe mailing list
> Haskell-Cafe at haskell.org
> http://www.haskell.org/mailman/listinfo/haskell-cafe
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.haskell.org/pipermail/haskell-cafe/attachments/20080505/ba460b37/attachment.htm
More information about the Haskell-Cafe
mailing list