[plt-scheme] Reexpansion of modules
--- Matthias Felleisen <matthias at ccs.neu.edu> wrote:
> Someone else recently wrote to the list about writing
> code that
> required serious performance improvements. I believe it
> had to do
> with number crunching. If it was _you_, could you remind
> us of the
> problems? -- Matthias
It was me. My problem boils down to this: calculate in the
Viterbi path of a hidden Markov model (HMM) as quickly as
possible. (My actual situation has some wrinkles, but the
slow part is essentially this, and the Viterbi algorithm
for HMMs is well known). I don't have statistics yet (I'm
working on collecting them) but I expect Scheme arrays to
be significantly slow than C arrays (I expect ~4-10 times).
When looking at runtimes in the order of hours or days
these constant factors make a significant difference. The
additional wrinkle is that I want clear hackable code --
this is part of my research so I'll be making many changes
to it as I explore different algorithms.
N.
Email: noelwelsh <at> yahoo <dot> com noel <at> untyped <dot> com
AIM: noelhwelsh
Blogs: http://monospaced.blogspot.com/ http://www.untyped.com/untyping/
__________________________________________________
Do You Yahoo!?
Tired of spam? Yahoo! Mail has the best spam protection around
http://mail.yahoo.com