On Thu, Oct 14, 2010 at 10:22 AM, Jay McCarthy <jay.mccar...@gmail.com> wrote: > I wrote mine in Racket then converted to Typed Racket. Attached is a different version of Jay's original code, ported to Typed Racket. There are 4 changes to the runtime code:
1. use `vector-set!' and `vector-ref' instead of `dict-update!' 2. use `in-indexed' instead of `in-dict' 3. use `compose' with only two arguments, instead of N. 4. an assert to state that the result of `string->number' is an exact integer I also had to add a type for `compose', now pushed. There are also many type annotations, which avoid the rest of the changes Jay made. Hopefully, we'll have time to add the types for dict functions on vectors in the near future. `compose' will remain a 2-argument function, though, unless someone has some very clever ideas for extending the variable-arity polymorphism work. As for timing, here's what I get. For compile times: [sa...@punge:~ plt] time raco make benford-ut.rkt real 0m0.458s user 0m0.372s sys 0m0.084s [sa...@punge:~ plt] time raco make benford.rkt real 0m1.867s user 0m1.708s sys 0m0.120s Approximately a factor of 3 in compilation time. I think Jay's numbers aren't right, since there's a big difference between the 'real' and 'user' times. For run times, on this benchmark: (define lst (build-list 100000 (λ (v) (random 11)))) (collect-garbage) (collect-garbage) (collect-garbage) (time (sort-em (sk lst))) [sa...@punge:~ plt] racket benford.rkt cpu time: 296 real time: 295 gc time: 8 [sa...@punge:~ plt] racket benford-ut.rkt cpu time: 292 real time: 292 gc time: 8 With minor noise, I think they're basically the same. Unfortunately, none of the optimizations we're doing have any impact on the computation going on here, so it's not that surprising. Vincent and I have some ideas for optimizing this code, so stay tuned. -- sam th sa...@ccs.neu.edu
benford.rkt
Description: Binary data
_________________________________________________ For list-related administrative tasks: http://lists.racket-lang.org/listinfo/users
