Poker-AI.org

I'm at the beginning phase of building a bot that's supposed to play on high-stakes play money tables, 6 to 9-max. The players there range from really bad to somewhat competent.

An important requirement is that decisions should take 2 to 3 seconds or less.

Honestly, I'm somewhat clueless about this, can't decide what is the right approach. Most of poker AI resources focus on heads-up, not games where 5 players seeing the flop is a normal occurrence. Are there any previous posts on this forum I should read? I'm especially interested in simple, baseline algorithms, that work reasonably well full-ring against weak opposition. Something that is not too complex and can be improved on iteratively.

Right now, I have two approaches in mind. The first is a rule-based algorithm based on how I play. The second is a theory-based (GTO) algorithm based on the strategy taught in the Upswing course. However, this could be harder to implement, partly because I'm not sure how to adjust that strategy for multiway pots and very atypical opponents.

Thougts?

Thanks

If you know how to play poker well I'd suggest you use your expertise to develop an expert system. But with a twist. Don't write endless rules about how to play certain hands.Instead think deeply about what really underlies your decisions and and use that instead, so the number of rules is smaller.

Yeah, I think Spears is right. It's called a rule based system I think.
The problem with a GTO approach where the bot is self learning.. is that you need a super computer that is like.. I don't even know how much it would cost.. too expensive. The HU bot that beat Doug Polk was a super computer and even then it had to spend about 20-30 seconds every decision.

By GTO approach I didn't mean self-learning AI (what Libratus and other heads-up bots use), but simply implementing the advice taught by Doug Polk:

Split your range to strong, medium, weak and weakest (relatively to opponent's range). Bet strong and weak, check / call medium and fold weakest. And adjust to exploit unbalanced opponents. The problem is how to decide what's strong relative to opponent's

If you know villains strategy then you can calculate the probability of villain holding each hand (his range). From that you can find your strength.

There's a lot of ways to approach GTO, I think you need to read up on that but I think an easy way to at least start off with approaching GTO is to not care about villains ranges at all. Just focus on making your own ranges balanced and unexploitable. Begin by looking up Mathew Janda on GTO 3/4/5betting. And learn how to analyze your database so you can check which opening hands are profitable to open, maybe you are losing money opening ATs UTG? Then go adjust the opening ranges for your bot.