TheNuts lab
From PokerAI
TheNuts is a pokerbot lab that came with the setup announcement quoted below. They have repeated this announcement in several different forms in various forums, however there is no knowledge of their success of failure to achieve their goals.
Team name: The Nuts (working name)
Project name: No Limit Poker Bot
Brief description: A computer playing world class Poker (i,e., No Limit Texas HoldEm) is considered by many to be the ultimate in Artificial Intelligence. It has never been done before (at least with any public awareness), despite very substantial efforts.
We have an existing team of four people developing such a No Limit Poker bot. The bot is based on Bayesian theory. For each opponent we assign a probability that they hold each possible hand; each time our opponent takes an action we multiply that prior probability times the conditional probability that the opponent would have bet or called with each hand given their particular situation. When it is our turn, we simulate the hand out resulting in an expected value of a call and raise. To do that we deal cards to each opponent according to the probability distribution we have assigned to them for each possible holding. Then we simulate the hand out to the finish based on a model of how our opponent plays each possible holding.
Over the last 6 to 8 months, we have completed mapping the entire bot (either in spreadsheets, C++, C, pseudo code, or word documents). The key now is implementation!
We believe this bot will be extremely profitable once completed. It has been designed to play $2/$4 to $10/$20 No Limit ring games and $50 to $200 entry Sit & Go tournaments. Again, the thinking is complete for tournaments, and it also uses a Bayesian approach.
Opponent modeling is the biggest challenges for us, as it is in any Bayesian bot. It is also complete in terms of thinking but it is not coded. We need to record a lot of data about each decision within each hand. Data collection, management, and analysis are key aspects of this bot.
We are confident we have figured this out; however, the quality of our implementation will be a primary driver of our success. Skills sets that will help here include knowledge and expertise with Poker Tracker, PostGres, SQL, statistics, math, logic, linear regression, and perhaps neural networks (though we aren’t in any way depending on neural networks, which we regard as a high risk implementation approach).
Target aim: We expect to complete the bot in 3 months (which probably means it will take 5). Then we expect to improve it indefinitely.
Compensation: Each team member will own the bot. Compensation will come from using the bot to play poker. We expect (and hope) to be making $3 - $8 per table per hour. One computer can play 5 tables at a time. $5 * $5 * 20 hours / day = $500 per day with one computer (and three poker accounts). If we are successful this will be exceedingly profitable for everyone.
Technology: The bot is being coded in C++ in a dll. We use www.winHoldem.net to help scrape the screen and push the buttons automatically (at least initially). We will use autoIT to develop a system whereby the computer can pick a good table, sit down, click the necessary button’s to start playing, and leave from the table after a few hours, take a break, and start playing again. We may incorporate neural networks (at least we will experiment with them) to improve the opponent modeling.
Talent needed: Most Valued Contributions from New Members: Object oriented C++ coding experience and knowledge, high energy & motivation, and statistics/database expertise (i.e., PostGres, SQL, statistics, math, logic, linear regression, and perhaps neural networks).
Valued Contributions from New Members: AutoIT skills, NL HoldEm expertise (ring & SNG), experience w/ WinHoldEm, and experience w/ Bayesian approaches and conditional probability.
Team structure: I’m the project manager. I can code in C, but I’m not very good at it. I’m a Harvard MBA graduate; have worked in management consulting and in private equity for many years. I left a very high paying job to pursue this opportunity and I do it full time. Other Members: We have 3 other members each of whom is very good at C++, know poker very well (as do I) and a few know autoIT.
Website: There will never be a web site. If you are interested I can PM you a personal web site describing me better.
Contacts: Please email interest with a summary of your background, capabilities relateive to our needs, and how you can be help. (thenutsproject@gmail.com)
Previous Work by Team: My teams have developed profitable bots in the past. However, this is an entirely new approach based on many lessons learned. While we are confident, we do recognize that we are trying to do something that has probably never been done before.
Additional Info: To be clear, we are not developing a game per se. We are developing a program to play online poker just like any human (but hopefully much better!)
Feedback: ENCOURAGING ONLY
See also
