headers
Douglas Bagnall | 12 Apr 2009 06:19
Picon
Favicon

update, X errors, exhibition

hello

I have been intermittently tinkering with tetuhi.  Some months ago I
removed the GSL dependency and ought to have improved the pseudo-random
qualities of the perceptrons by including dSFMT[1], which led me to
improve some of the ways those numbers were used.

More recently I found that tetuhi was crashing with XIO error messages
when it forked to work in the background.  Previous episodes of this
kind have been caused the child processes making noise on (or attempting
to close) the X sockets, so I assumed some update in pygame or X or SDL
or something had introduced new sensitivity, and tried a haphazard
variety of X swaddling techniques.  With the sockets closed with a dup2
to a "[arw]+" file, the error message to changed to "XIO:  fatal IO
error 0 (Success)" but the children still died.  Then with print
statements I found it was always crashing in a routine related to the
new perceptron randomness code mentioned above, so perhaps X is innocent
except of aiding and abetting bugs by disguising their origin.  If
anyone cares to look the code in question is in the "broken" git
branch[2] (which is not to suggest the master branch is working: it just
contains less debugging code).

So that sounds like a problem just about to be solved (i.e. it is my
stupid perceptron's fault), and you are wondering why I am relating it
at this point and not waiting for the triumph.  The reason is that Te
Tuhi Video Game Machine, the hardware/artwork that initially ran the
code is being shown again, this time at the NZ Film Archive in
Wellington, New Zealand[3].  The exhibition opens at 5:30 on Wednesday,
and if any of you are in this city, please come along and enjoy the free
drinks and food.  The software on the machine is exactly that that was
(Continue reading)

Permalink | Reply | 0 comments |
headers
Douglas Bagnall | 18 Aug 2008 07:15
Picon
Favicon

r.carter@... requires approval

framebuffer problem

Robert Carter <r.carter@...> wrote the following
but it
got caught up in the mailman post approval system so you won't have
seen it yet:

> Hi Everyone,
>
> I've been able to compile tetuhi, but I have a problem with the
> frame buffer. Initially directfb did not detect the framebuffer
> device /dev/fb0. I added a kernel option like 'vga=791' to activate
> the fbcon framebuffer on startup. This allowed tetuhi to run, but
> with unusual results. The display shows a 'comb' of horizontal scan
> lines which are moving. It looks like the software is working - it
> gives output on STDOUT that looks reasonable. I'm testing this on
> an intel imac with a 'cinema' width display. Perhaps I need to set
> something to tell the software how many bytes per horizontal line,
> or some screen size related environment variable?
>
> Rob

It sounds like a PyGame/SDL problem, but beyond that I have very
little idea.  The SDL_VIDEODRIVER environment variable might be
useful.  There's some explanation here:

http://www.pygame.org/docs/ref/display.html#pygame.display.init

Also the set_mode function further down that page might be relevant.
(Continue reading)

Permalink | Reply | 0 comments |
headers
Douglas Bagnall | 18 Aug 2008 07:13
Picon
Favicon

r.carter@... requires approval

framebuffer problem

Robert Carter <r.carter@...> wrote the following
but it
got caught up in the mailman post approval system so you won't have
seen it yet:

> Hi Everyone,
>
> I've been able to compile tetuhi, but I have a problem with the
> frame buffer. Initially directfb did not detect the framebuffer
> device /dev/fb0. I added a kernel option like 'vga=791' to activate
> the fbcon framebuffer on startup. This allowed tetuhi to run, but
> with unusual results. The display shows a 'comb' of horizontal scan
> lines which are moving. It looks like the software is working - it
> gives output on STDOUT that looks reasonable. I'm testing this on
> an intel imac with a 'cinema' width display. Perhaps I need to set
> something to tell the software how many bytes per horizontal line,
> or some screen size related environment variable?
>
> Rob

It sounds like a PyGame/SDL problem, but beyond that I have very
little idea.  The SDL_VIDEODRIVER environment variable might be
useful.  There's some explanation here:

http://www.pygame.org/docs/ref/display.html#pygame.display.init

Also the set_mode function further down that page might be relevant.
(Continue reading)

Permalink | Reply | 1 comment |
headers
Matt Fischer | 27 Jul 2008 20:48
Picon

nnpy module?

I downloaded the tetuhi source tonight and have been working through the missing dependencies that I had.  I was able to compile finally, but when running, I now see this:

ImportError: No module named nnpy

I'm pretty new to python, but I did poke around some and did not find a module named nnpy, any hints?

mfisch <at> localhost:~/tetuhi$ ./tetuhi ~/Desktop/beach.jpg
File /etc/tetuhi.conf not readable
File /home/mfisch/.tetuhirc not readable
File ./config/tetuhi.conf not readable
File ./tetuhi.conf not readable
found 1 CPUs
Using 1 processes
 children have 6 secs, parent has 8
Traceback (most recent call last):
  File "./tetuhi", line 176, in <module>
    from vg.maker import GameMaker
  File "/home/mfisch/tetuhi/vg/maker.py", line 26, in <module>
    from vg.game import Game
  File "/home/mfisch/tetuhi/vg/game.py", line 30, in <module>
    from vg import actor, weights
  File "/home/mfisch/tetuhi/vg/actor.py", line 23, in <module>
    from vg import config, utils, weights
  File "/home/mfisch/tetuhi/vg/weights.py", line 37, in <module>
    import nnpy
ImportError: No module named nnpy

_______________________________________________
Tetuhi-vgs mailing list
Tetuhi-vgs@...
http://lists.nongnu.org/mailman/listinfo/tetuhi-vgs
Permalink | Reply | 0 comments |
headers
Douglas Bagnall | 27 May 2008 04:45
Picon
Favicon

release of 0.85


After much delay (attributable to the birth of a child), version 0.85
is out.  The best new thing is (partial, at least) OLPC XO support.
It runs well but is un-sugarised and several configuration variables
need tuning for the slower machine.  To use the XO camera you need the
-x option.

A fuller Changelog looks like:

  - Yaml import uses either PySyck or PyYaml, depending what is
    available.
  - Improvements to Makefiles, so the modules build on more systems.
  - Adaptation for OLPC XO (incomplete, but it runs).
  - More flexible (fully executable) config files, and option to
    dump current config values.
  - Various fiddly changes and sort-of clean-ups or improvements.

Thanks to John Knight and Eric Piel for finding problems with the
build system.  Tetuhi is now much more likely to compile, and should
even work on 64 bit machines.

Because of the noob difficulties I had in cross-compiling for the XO,
I have included binaries of the C modules (under binaries/olpc), so
other novices can get going quicker.

The tarball is as always available from

https://sourceforge.net/project/showfiles.php?group_id=217385

and you can get the git tree directly from

http://git.savannah.gnu.org/gitweb/?p=tetuhi.git

Douglas Bagnall
Permalink | Reply | 0 comments |
headers
Eric Piel | 13 Mar 2008 22:41
Favicon

Some fixes for x86_64

Hi,

This is a quick message to summarize what I needed to change (against 
the git tree) to run tetuhi on my x86_64 machine:
* Compilation must be done with -fPIC. This implies not only changing 
the COMMON_OPTS variables, but also running "python setup.py build_ext" 
with "CFLAGS=-fPIC LDFLAGS=-fPIC"
* "pyserf_helpers.c" is not found in nnpy.c, it must be changed to 
../lib/pyserf_helpers.c
* The main difficulty: collect_entropy() of vg/cblobdetect.py should 
return a list of integers, not of longs (otherwise, overflows appear) so 
line 315 should be:     return list(array.array("i", e2))

The last two points could probably be directly fixed in the source, as 
described.

Hoping it helps some people :-)
See you,
Eric
Permalink | Reply | 1 comment |
headers
Douglas Bagnall | 29 Feb 2008 09:45
Picon
Favicon

publicity message

I've sent this, or close variations, to one or two places:

Announcing the public release of Te Tuhi Video Game System

Te Tuhi is not a game in itself; rather it creates games based on
arbitrary images that it is given.  To use it, you draw a picture of
the game you want to play, and it will give you the game that you
really drew.

The software was originally written for an exhibit at Te Tuhi Centre
for the Arts in Manukau City, New Zealand, from which it borrowed its
name.  That show ended on 10 February 2008, at which point the
software was released under the GPL.

It is written primarily in Python and C. Experimental development is
encouraged, with the system being sufficiently modular to allow people
to mix their own combinations of image parsing and game evolution
implementations.  A version of the software is being developed for the
OLPC XO.

The project has a home page at http://halo.gen.nz/tetuhi/code.html, a
git repository at http://savannah.nongnu.org/git/?group=tetuhi, and a
mailing list at http://lists.nongnu.org/mailman/listinfo/tetuhi-vgs.

A tarball of version 0.83 (the first public one) can be found at
http://sourceforge.net/project/showfiles.php?group_id=217385 but
sampling from GIT is recommended.

Te Tuhi Video Game System was originally written by Douglas Bagnall.
Permalink | Reply | 0 comments |
headers
Douglas Bagnall | 29 Feb 2008 09:41
Picon
Favicon

get this list started.

Welcome to the tetuhi mailing list.

The main purpose of this message is to trigger the creation of the
archive pages.  A 404 error looks even more lonely than an archive with
a single message as vacuous as this one.

douglas
Permalink | Reply | 0 comments |

Gmane