Thesis about Tzaar and AI

Modérateurs : minervois, Piano, lynkowsky, alesk

Re: patch for --output

Message par Paulie » 30 Sep 2012 16:25

alihv a écrit :Thanks for the great work and congrats!

I've added --output option to prevent overwriting input files:
http://pastebin.com/bZB5EcXT


That is useful, thanks! I'll put the patch on the repository on Google code site soon (with some other stuff).

alihv a écrit :also, a bug report: -a 1 segfaults on this position: http://pastebin.com/7F9gb2iu


Thanks for reporting. There is a bug in the simplest Alpha-Beta implementation (which is not used by the robot). I'll push the repair with your patch.

alihv a écrit :I've written a plain UCT algorithm and I'll test it against your bot.


Great! I'm not able to write a UCT algorithm :-) I think that it is quite hard, isn't it? Also, I would like to see your code.
Paulie
 
Message(s) : 48
Inscription : 09 Jan 2012 1:15
Localisation : Czech republic (Praha/Strakonice)

Re: Thesis about Tzaar and AI

Message par alihv » 01 Oct 2012 2:35

Surely you're able to write a UCT algorithm! The only problem is that if something's messed up, the bot just plays badly and you don't know why. Fortunately with plain UCT (no RAVE or other crazy stuff) there are not too many opportunities to screw up.

The code's in Scala, OK? It's not releaseworthy, but here goes anyway:
Code : Tout sélectionner
git clone git://gitorious.org/uctprove/uctprove.git

But check out fuego for a much better UCT implementation (and in C++). Somewhere I have a version of fuego-1.1 that compiles.

A 100-game match has just been finished:
white: UCT with 10'000 simulations per half-move, after a move the corresponding subtree persists
black: PauliebotBeginner
45:55
alihv
 
Message(s) : 9
Inscription : 28 Mars 2012 12:57

Re: Thesis about Tzaar and AI

Message par SlowBrain » 14 Oct 2012 10:12

Very interesting reading!The strategy used by the bot might be a clever idea, although I must think more about it to have a clear verdict if that is how humans should play.

After 15 games against PauliebotUnbeatable I have 8 wins and 7 losses. The bot is very strong!
SlowBrain
 
Message(s) : 35
Inscription : 07 Nov 2010 20:07
Localisation : Stockholm, Sweden

Re: Thesis about Tzaar and AI

Message par Paulie » 14 Oct 2012 15:53

SlowBrain a écrit :Very interesting reading!The strategy used by the bot might be a clever idea, although I must think more about it to have a clear verdict if that is how humans should play.


Thanks. I would write some more observations in the strategy part of the thesis now :) And the robot is not playing exactly according to the strategy, it is sometimes losing a stack of 3 or 4. The evaluation function is the magic :)

SlowBrain a écrit :After 15 games against PauliebotUnbeatable I have 8 wins and 7 losses. The bot is very strong!


Wow! I didn't expect this!
Paulie
 
Message(s) : 48
Inscription : 09 Jan 2012 1:15
Localisation : Czech republic (Praha/Strakonice)

Re: Thesis about Tzaar and AI

Message par degeus » 07 Mai 2013 18:30

Nice :-) Anyone know how to make this into an Android app ;-)
Avatar de l’utilisateur
degeus
 
Message(s) : 11
Inscription : 12 Déc 2011 10:12

Re: Thesis about Tzaar and AI

Message par Paulie » 07 Mai 2013 21:51

Hi!

degeus a écrit :Nice :-) Anyone know how to make this into an Android app ;-)


Thanks :-)

What do you mean by Android app?

Btw, a new version of the thesis and also much shorter article can be found on http://kam.mff.cuni.cz/~vesely/tzaar/
Paulie
 
Message(s) : 48
Inscription : 09 Jan 2012 1:15
Localisation : Czech republic (Praha/Strakonice)

Précédent

Retour vers Tzaar

Qui est en ligne ?

Utilisateur(s) parcourant ce forum : Aucun utilisateur inscrit et 1 invité

cron