Worksheet 2

Honours Project Worksheet — #2

My research question will be "Can a no-limit Texas hold'em AI learn from other players and change strategy accordingly?"or something to that effect.

The focus of the project will be to build an AI that can use betting strategies to out smart and out play opponents. When the final program has been built, it will, at first, play against other AI's (and if time allows, have the AI different "personalities") and then I wish to make it be able to play against human players.

Possible expansions:

Have the AI predict an opponents hole cards
Teach the player what it has done wrong
(I doubt I will have time to make both of these possible, but it would be possible to do for future expansions outside the honours project)

Final Idea & Work sheet 1

Honours Project Worksheet – #1

1. State the topic of interest that you may wish to take forward into a project.
2. Indicate the issues surrounding this topic that you may wish investigate.

1. I wish to make an AI the can not only play Texas hold'em poker but adapt to how the opponent plays, using various betting strategies and advanced game theory.

The goal of this would be to build a program so that it could beat a good player and a bad player, and (most importantly) win the most money out of them. It should see how the opponent plays (ie. Do they bet a lot? Do they play a lot instead of folding?), and use this to their advantage. For example, if the enemy is a "maniac" (Plays a lot of hands and constantly bets, raises and re-raises), then the AI should start playing only the top hands and only bet/raise when it is totally sure that it has the best hand.

2. The main problem I have at the moment is that I am not sure what to use as a framework. I don't want to spend a lot of time building just the set-up (ie. the dealing of card, the graphics etc), as this will take away from time I have to make the AI as good as possible.

The maths of working out good betting strategies is in nearly every single poker book and since I am an avid player myself, I have personal experience on which strategies work and why.

I also wish (If time allows), to have the AI "guess" the opponents hole cards, like what a professional would do. In reality, pro's use the information that has been given by the previous betting on each different betting level and puts them on a range of possible hands and the more bets are made the more this range can be shortened. It also would take into account the type of play the enemy is, and use this to determine the most likely cards they have.

As a side note, I do not wish to make the computer cheat in any way, shape or form. If everything goes well, it should only play with the hand it has been given and the bets.