Hi there!
I wanted to create a pokerbot some years ago, but never really got to it. Having quite an understanding of the game of poker, and a decent amount of experience playing it, and a (theoretic) background in AI and ML, I am just not enough of a programmer to just sit down and get to it all by myself I guess.
------- where I come from -------
Taking another dive into the topic just recently (triggered by Dahls FLHU ANN machine) I found the whole poker AI/bot scene seems to have evolved quite well, and there even are several "commercial" version out there now. I guess most of them are a scam, but since I believe that bots can at least beat micro and small stakes I can imagine that one of the offered solutions works well enough.
For my understanding there are 2 parts to the solution: - poker client interaction, and - the game logic itself
For the first part the options are screen scraing, dll injection, or even network traffic analysis from what I read. For the game logic there there are numerous options, e.g. expert systems, game tree search, MC simulations, game theory approaches, evolutionary algorithms, case-based reasoning ect. What I always had in mind was an ANN, probably trained by an set of hands played by a good player. A good selection of input to the ANN would be crucial. Maybe an combination of expert systems preflop, and ANN postflop would also do. On top of hand strength and pot-odds, I assume position, an opponent model, the betting sequence of the hand, and whatever output it gives a bit randomized to increase unpredictability to be among the most important features. But this could also way to naive, due to the lack of experience I do not know better at the moment. If someone wants to discuss that or help me out feel more than welcome.
--------- what I wonder at the moment --------
Having all that said, my current interest is in the openholdem platform as the part doing the client interaction. Is it "stealthy" enough to not get detected? I would imagine that poker platforms also know this framework and can therefore easily spot it?
The net question would be: how good are the AIs that are available for it? I have seen that some come along with the package, but there is a whole marketplace to get numerous AIs from, most focused on a special game (e.g. SnG, MTT, 6max NLHE ect). Is any of them good enough to actually beat real money (small stakes) games? And if so, which one would you consider very strong? Having looked into the code-base of some AIs it seems that they are all build as expert systems, or do I miss something here?
I am happy if anybody wants to discuss that matter, or even wants to team up for a deeper dive into the topic.
|