people playing: 48, servers online: 16, games played: 142,876,560 Top Player: Ceytune

  Toribash Community » General » Discussion
Register FAQ General Rules Community
Reply
 
Thread Tools Display Modes
Old Dec 14, 2018   #31
missuse
Nobody fucks with the Jesus
 
Red Belt
Join Date: Oct 2008
Posts: 13,605



Originally Posted by notchnick View Post
Sorry If I will have mistakes. I think AI for toribash will be a good idea. It will work like a chess cause in toribash you click on joints. It's not hard movement. So I think it's will not be so hard

yes but in chess you have set rules for every game in tb rules change per mod - gravity, dm trheshold, no DM, fractures. But yea I agree it can be done with sufficient effort.
wishful eyes deceive me
missuse is offline   Reply With Quote
Old Dec 14, 2018   #32
Divine
8 - 6
 
Gamecock Feathers Belt
Join Date: May 2014
Posts: 1,152



Will if you're talking AI that learns from experience it would be pretty tough seeing as toribash is an incredibly complicated game and it would take a lot of games for it to learn anything. You would need a large community effort to teach it.

You might be able to pit it against itself though then it runs into the possibility of creating odd and ineffective habits.

It is doable though would be incredibly difficult for a small team or as a solo effort.
Divine is offline   Reply With Quote
Old 1 Week Ago   #33
Miffylin
Junior Member
 
White Belt
Join Date: Sep 2012



Sorry to necro(?) this thread, but a small shameless self-plug: I released a machine learning interface for Toribash a while ago , and over these past months I have been using it and fine-tuning. I am going to release 1.0.0 in couple of days.


In short: Toribash is very suitable for such a learning env! I am able to run 16 instance of Toribash on Linux machine without bigger issues (minus maybe some slowdown), and they run fast enough for decent experiments. However applying the OpenAI/DeepMind type of brute-force approach and going pure self-play seems to require a ton of training, even for game like Toribash, to produce anything sensible.

Last edited by Miffylin; 1 Week Ago at 10:36 PM..
Miffylin is offline   Reply With Quote
Old 1 Week Ago   #34
Azraeel
Member
 
2nd Dan Black Belt
Join Date: May 2018
Posts: 53
Clan: Broken



to comment on this thread. Lets say you put the AI in the fight with a player like diamond. This ai is gonna learn quickly. therefore i think it would be a cool idea. i'd think theres a lot of possibilites of how you can use the ai too **In events ofc***
Manifest is my family. Manifest is my dreams and work. Manifest is special to my heart.
Azraeel is offline   Reply With Quote
Old 1 Week Ago   #35
Mocucha
 
7th Dan Black Belt
Join Date: Jan 2010
Posts: 4,292
Clan: Evolution






Maybe in the future, that kind of AI would be implemented in Toribash Next ;)

My Youtube channel : Mocucha Toribash

Mocucha is offline   Reply With Quote
Old 1 Week Ago   #36
Sora
Legendary Troublemaker
 
10th Dan Black Belt
Join Date: Dec 2016
Posts: 4,569
Clan: Liquor



Originally Posted by Mocucha View Post



Maybe in the future, that kind of AI would be implemented in Toribash Next ;)

probably not xd
Google deepmind costs more than 1million $/ year just saying
Team Lenshu|Anime United
Sora is offline   Reply With Quote
Old 1 Week Ago   #37
Azraeel
Member
 
2nd Dan Black Belt
Join Date: May 2018
Posts: 53
Clan: Broken



Originally Posted by Sora View Post
probably not xd
Google deepmind costs more than 1million $/ year just saying

Lol
Manifest is my family. Manifest is my dreams and work. Manifest is special to my heart.
Azraeel is offline   Reply With Quote
Old 1 Week Ago   #38
Miffylin
Junior Member
 
White Belt
Join Date: Sep 2012



Originally Posted by Sora View Post
probably not xd
Google deepmind costs more than 1million $/ year just saying


Quite so, and especially when it comes to the Dota2 and SC2 bots these big companies have been presenting, those experiments literally used millions of dollars for cloud computing services.


I have ran quick experiments with various self-play techniques (see Dota2 and SC2 bots) on Toribash with simple rules (game length 500 frames, 10 frames per turn, disqualification enabled, no dojo), and after couple of weeks of non-stop training and ~6M games the AI does not seem too good. That is not to say it is not feasible, but it will require finding good methods and possibly quite a bit of fine-tuning to get it right
Find me on Toribash Discord if you wish to get in touch. I may miss PMs here.
Miffylin is offline   Reply With Quote
Old 1 Week Ago   #39
Azraeel
Member
 
2nd Dan Black Belt
Join Date: May 2018
Posts: 53
Clan: Broken



Originally Posted by Miffylin View Post
Quite so, and especially when it comes to the Dota2 and SC2 bots these big companies have been presenting, those experiments literally used millions of dollars for cloud computing services.


I have ran quick experiments with various self-play techniques (see Dota2 and SC2 bots) on Toribash with simple rules (game length 500 frames, 10 frames per turn, disqualification enabled, no dojo), and after couple of weeks of non-stop training and ~6M games the AI does not seem too good. That is not to say it is not feasible, but it will require finding good methods and possibly quite a bit of fine-tuning to get it right

Hope you find a way then
Manifest is my family. Manifest is my dreams and work. Manifest is special to my heart.
Azraeel is offline   Reply With Quote
Reply


Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off

Forum Jump


All times are GMT +1. The time now is 10:58 AM.




Powered by vBulletin® Version 3.8.2
Copyright ©2000 - 2019, Jelsoft Enterprises Ltd.
ragdoll fighting game physics fighting game ragdoll fighting physics funmotion joints martial arts karate pc mac free game turn based game