[Icfp04-discuss] Scheme, Ruby ?

[Alejandro Forero Cuervo]

Hi.

> Why  you decided  to generate  bitmap images  instead of  using
> vector  graphics ?  Have not  found apropriate  vector graphics
> library for Scheme ?

We used  PNM format because we  thought it would be  the solution
that would take us the least  time to implement (we had used that
format relatively  frequently in the  past).  It took  around 100
lines  plus a  few definitions  for  the actual  images that  the
render is made of.

Another option  we considered  was generating SWF  (Flash) files.
This would  probably allow us  to control the  visualization more
directly (such as interactively changing  its speed) but it would
probably have taken more time.

We  can  generate  Flash  files  with  SWFTools  from  our  PNMs.
Animations stored in this format  have about 1/10th the size they
have when  stored as  GIF files  (created with  ImageMagick).  Of
course, this  wasn't a  concern during the  contest.  There  is a
sample in:

    <http://www.crazyrobot.net/anim.swf>

> Have   you    tried   "Scheme    Widget   Library    (SWL)"   [
> http://www.scheme.com/contributed.html   ]  for   visualization
> ?   It    can   be   used    with   Petite   Chez    Scheme   [
> http://www.scheme.com/petitecs.html ]

I have  not tried it.   I don't know  about the other  members in
our  team, but  not  having  free (libre)  access  to our  Scheme
implementation  would  be  a  price  too big  to  pay  for  being
able  to use  that library.   Of  course, porting  it to  another
implementation is always an option.

> > Our simulator  took around  30 secs to  run a  100.000 rounds
> > match.  We didn't spend any time optimizing it.
>
> It would be  interesting to profile your simulator  to find the
> code  consuming  most  of  the time  in  a  single  round. What
> structure you used to represent ant world in Scheme ?

It can certainly be optimized.

; markers is a bitmask.  The first 6 bits are for the
; red ants, the second 6 bits for the black ones.

(define-record mapinfo terrain food ant markers)

; type is a boolean, with #t for red, #f for black
; food is a boolean (#t means it has food).
; alive is a boolean

(define-record ant x y dir food type state rest alive)

We  use a  global  variable *map*  with a  vector  of vectors  of
mapinfo records.

Things that I would optimize:

- Movement  and   senses.   Some   (case  ...)  forms   could  be
  eliminated.

- Parsing  tokens directly  as  they are  read.   There are  some
  computations that are  done during the execution  of the rounds
  but could  be done  as the instructions  are read.   This would
  involve rewriting  read-ant-state and the *-exec  functions.  I
  might do that latter and check if it actually improves things.

- Compiling the  code with -optimize-level  3 -usual-integrations
  -unsafe  -block   -fixnum-arithmetic  -lambda-lift   and  other
  similar arguments.

I might  apply some  of these  changes and  see if  they actually
improve things.  I know, I should use a profiler first.

The reason we didn't optimize  our simulator is mostly because we
were  not  very  interested  in doing  genetic  programming.   We
decided  time  spent writing  ants  was  better than  time  spent
optimizing our  simulator (given that  30 secs was a  decent time
for it to take).

> (I couldn't access your repository at:
> http://anonymous:@svn.afc.no-ip.info:8000/svn/home/src/icfp/)

Sorry, it probably means my machine was offline (as it gets every
night (GMT-5)).

Alejo.
http://bachue.com/alejo

---=(  Comunidad de Usuarios de Software Libre en Colombia  )=---                                                              
---=(  http://bachue.com/colibri )=--=( colibri at bachue.com  )=---                                                              
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: Digital signature
Url : http://lists.seas.upenn.edu/pipermail/icfp04-discuss/attachments/20040616/44a830c3/attachment.bin