How can I train a computer system to play poker?

Question

At first I have 5 cards randomly, of course. I can only change once, and I also taught the [program] poker rules in my system. My problem is how can I choose "I do not need this card or do I have cards?". I can change it myself, but the computer does not know. I think it might be difficult, but do you guys have any help?

Answer 1

Take a look at How I built a poker boxing working group (Parts 1 , 2 , 3 , 4 , 5 , 6 , 7 , 8 ), which has excellent information for getting started in poker botting.

In particular, see part 8 for details on how to evaluate the value of a hand. What you want to do is to statistically evaluate the cards in your hand and from there determine which cards are most likely to score a winning combination. You can discard cards that you do not need.

Answer 2

How about contacting the University of Alberta Computer Poker Research Group ?

Answer 3

Game theory and AI seem to be what you need to explore here.

Answer 4

If you just want to create a poker program, then the above suggestions are useful. If you want to create a system that learns to play poker, you need to use machine learning / inductive learning methods such as neural networks, decision trees, genetic algorithms. Samuel wrote an early checker playing program that learned to improve its performance.

http://en.wikipedia.org/wiki/Arthur_Samuel

There is also a website for computer training and games:

http://satirist.org/learn-game/

Answer 5

Be careful not to do too good a job, or you could lose a lot of money.

Answer 6

You will want to explore the Design Expert System . And apply it to your problem area.

If this seems overwhelming, start small. Create a rudimentary knowledge base about rules that will help your poker bot eagerly search for pairs (and 3 or 4 types). When this succeeds, add rules to try to build straight lines, as well as rules about when you need to support direct construction over pair searches, then when it works well, add flushes. Continuous and small increment is the key to the process.

Answer 7

Isn't it better to start with something simpler and work from there? How blackjack, which I believe will be relatively easy in a grand scheme of things?

Answer 8

Computer game by David Levy

alt text http://ecx.images-amazon.com/images/I/616X%2BHJmE7L._SL500_AA240_.jpg

Covers studs and draws. You will understand how much it will be by the time you finish reading.

Please note that the book is over 25 years old, but this is a good start.

Answer 9

What you want to learn is game theory and data mining. Top programs use a simulation of theoretical games to obtain tolerances for an expert system that is used to randomize a game so that it makes as little systematic errors as possible. If you search the Internet well enough, you will find complete implementations of some of these types of solutions, but often they lack the code that created them. The downside simply makes one of these decisions - it is that they do not make full use of the mistakes of opponents, and that is why most of today's research is directed on this topic.

There are several ways to generate theoretical solutions to the game using almost any conceivable AI algorithm.

Answer 10

Check the poker evaluator. http://www.codingthewheel.com/archives/poker-hand-evaluator-roundup Use monte carlo analysis to find which cards to upload to oreder to maximize expectations.