Poker-AI.org

hi folks,

first of all - sorry to see the old pokerai.org forum gone, what happened to it?

Anyway, so for the last 2 years I was flying under the radar a bit because I managed to start botting on PS and tried to move up from the micro stakes as far as possible. Together with a friend who actually plays poker himself I now managed to get to 50NL.

But I have to say there is something that deeply bothers me about all of this. Here we are, applying poker research and building something amazing (as I'm sure many of you are) - but we can't really show it to anyone. There is so much exciting stuff about building a poker bot, but there is also so much work that's really boring and has nothing to do with poker. Screen-scraping, adapting to changes in the clients, trying to stay under the radar, etc.

So I started looking into online competitions for poker, but many of them require you to send your source code or compiled program (no-go for me). Others, like computerpokercompetition.org, seem to be only annual events, and they also require your program to run on a slow machine.


What if there were something like an online casino for AI?

Something like a normal Casino with a public API for play money games (heck, maybe even real money if we want to).

The advantages to the existing AI competitions that I see:

* Run our bots 24/7 not only heads-up but in full-ring games against others.
* Be actually able to test your opponent model (!)
* It would allow researchers and newcomers to quickly get enough data to see how profitable their bots are.
* Higher-stakes poker botters could show off their bots without giving away their anonymity or any of their source code.
* Fame and glory of having your bot on top of the high score (until you get beaten, haha)

What are your thoughts about this? If there is enough interest, I would be happy to start working on something like this. I'm really curious how my bot would perform against others. Aren't you?

cheers,
Chris

Hey. I like your idea but I am not really sure how useful it would be. I have been doing this for nearly a decade and from that experience I think its better to test bots against real players. Humans play differently..alot differently then a bot. It is not expensive to put your bot in micro limits and see how it reacts. You will lose money but the knowledge gained is worth the price. Then its just a matter of going over the HH files and adjusting each bad fold, bad call etc until it no longer makes any mistakes worth mentioning. This process could take months (when I first started in FL it took a year) but its worth it.

Then there is the type of game. I think the true test of a bot AI is NL HU as its a game of marginal holdings, position and bluffing. In general your not going to get away with that in 6 max or full ring games so its way less scenarios to program for.

I spent 2 years on NL HU and I am still not happy with it and constantly making adjustments. It is definitely the toughest bot Ive ever had to write. The other thing about HU is that I learned more about game theory then all of the other games combined. PLO is where my attention is now but that is another story.

But with all that being said I think NL or PLO 6 max Zoom is the ideal environment for a bot. Zoom/Rush is the greatest invention for bot runners since PPT..

EDIT. I should say that 100BB+ HU NL is a true test of of poker AI. But if you want to make money then I think 6max Zoom is the best game to be in with less difficulty in the AI requirements.

Absolutely. And I would never suggest otherwise. But I'd like to think of the poker room as an alternative to the poker test bed. It's way easier to get up and running against an online service that's always up and populated with bots from different developers. And it would be awesome to see how your own bot will do against the better bots out there. Also out of curiosity, not just to improve things.



Not everyone is focusing on HU. Myself, I run a full-ring FL bot and try to keep it away from HU as much as possible (sometimes players leave the table and there is nothing I can do about it as sitting out each time would give too much away). I'd love to see how my bot performs against other bots in a ring game.


I started with NL micro stakes 5 years ago. Moved over to FL as it was definitely easier to go a little bit higher stakes (definitely a more mechanic decision making).
never tried zoom/rush variants yet but it's definitely on my list - must be a huge +EV

FL ring? You can still make a living in FL? Where?? I have a fully tested FL 10max AI that I stopped using in 2006. It won a ton until the game died off and NL took off. 2 years of development all to see the game die. At the end of FL I was barely able to find 2 decent games even at peak times. During the hay day of FL (at Paradise and then PartyPoker) you could play 8-16 tables at once making 2-3BB per hour each plus bonus clearing.

I did see a few games at FTP awhile back but not enough to put it into action again.

- I like the idea but you need to be sure it's worth your effort
- There is a plethora of sites that allow humans to play bots, but they are all bot specific. It might be useful to have a site that allows any bot to play any human too. And even nicer if that is open osurce.
- Personally, my old bots would just embarrass me, and I'm too busy to put much time into my latest ideas, so probably wouldn't participate in the short term
- http://www.poker-ai.org/archive/www.pok ... =a&start=0
- Playing enough games to overcome variance is problematic. You can't play duplicate because it's too easy to cheat.
- A competition of NE bots quickly gets boring, and not very relevant. SkillBet would be much more interesting http://en.wikipedia.org/wiki/Duplicate_poker

yeah. It would be a fun project for me, but I want to see if there is any interest.


Too bad

Thank you so much for pointing me to the koypetition! It's great to see that there is quite some interest in something like this. Too bad that they seemed to have their own agenda (they do poker analysis software, maybe they wanted to copy some ideas from the bots?)

do you know how many people eventually took part in the competition?


* depends on the difference between bot skills
* at an average bot response time of 2 seconds, a full ring game will take ~30 seconds, at 10 tables that would be 1200 hands per hour. Or close to 30k hands per day. that should be a good start, no?


yeah. also, the rules should be as close as possible to PS rules. no multi-accounting (but multi-logins to test different kinds of bots simultaneously), no collusion.


nice, didn't read about this one before. possibly the idea can be extended to include the other players/bots at the full ring table.

There used to be the old "Poki Poker Server" run by University of Alberta:

http://web.archive.org/web/200402020243 ... mes/poker/
http://web.archive.org/web/200408050044 ... /bots.html

but sadly it died a long time ago (and IIRC, the protocol only had limit games).

Juk

I did look into this and found the following open source but no longer being worked on.

It does come with a bot server and I believe a bot server can be run on a different machine/ network.

https://bitbucket.org/cubeia/cubeia-poker-agpl
http://forum.cubeia.org/viewforum.php?f ... b9f5d603fc

Screen shots
http://www.cubeiasocial.com/features/

Already have a working prototype for No Limit and Limit Poker, only the API needs to be finalized now - and I could use your input.

As a "side note": I realized that an API and "bot casino" like this would have economic advantages for all developers (see my ideas about "Poker Bot as a Service" here: viewtopic.php?f=26&t=2847 ). Would love to get your comments on this, especially from the "established" poker botters.

Requirements of the protocol / API
Here is what I think is necessary for bot / server communication:

[*] Bot sign in with respective game preferences. The bot needs to tells the server all the details about prefered games (NL? FL? HU? full-ring? tourneys?) and multi-table capabilities.
[*] Server needs to be able to request decisions from the bot
[*] Server needs to send hand histories of completed games to the bot (for opponent modeling)

So it seems that the protocol isn't actually that complicated if the server handles automatic seating according to the bots preferences. This means for the developer it's almost as easy as implementing the ACPC. However unfortunately the ACPC does not provide enough info to the bot (e.g. player names are missing, multi tabling not possible, and no hand histories etc.). We're probably looking at an extension of the ACPC.

Let me know about your thoughts, are any important requirements missing?
Cheers,

Did you guys ever look at the old pokerpro from the old winholdem?

SuitedAce is a real money online casino and has a lot of poker housebots running.

I would like also to know how profitable their bots are?

how hard would it be to write the server side code?

seems like a pretty straightforward project.

thanks

Related in a way:
http://theaigames.com/competitions/texas-hold-em

The housebots do not run anymore.

why?

I agree with opinion to test bot with real players. I'm playing poker for almost 5 years regularly. Most of that time here http://evenbetgaming.com/. I helped my friend to test two or three different bots, just for statistics. And the best result was after live players testing.

has anyone seen this. it relies on slack so as long as that's around (highly likely) it shouldn't go offline and there's no server costs for anyone

https://github.com/CharlieHess/slack-poker-bot

Nice.

Hi there. Yes, there is a play-money poker site specifically for online botting (https://aitestbed.com). It offers a clean API for interacting with the server and launching your bot to compete against other bots and humans in real-time.

The API is here --> https://metacpan.org/pod/Poker::Robot
Human interface --> https://aitestbed.com

I have two bots running on there now, Bender and Alice23. You can log-in and play against them anytime you want. I wrote Alice to mimic the way a I play poker, only better. She plays okay, but needs more work.

Any questions or interest, please feel free to contact me (I admin the site). I'd love to get more bots on there competing. The back story is that I wrote all this code a few years ago for a commercial project, but the financing fell through, so now I'm thinking about releasing it all as an open source project (maybe). Anyway, the API and protocol are all public, so bot away.

Very nice. When I get a free moment I will see about implementing some of my bots to the interface. How decent are yours at HUNL? Also, I noticed you only have your bots playing cash games. Any interest in setting up some HUNL SnG matches?