Poker-AI.org

Can anyone tell me what are the most important opponent stats when programming a heads-up bot?

There are probably 20-30 stats you will need. I would just google a typical HUD setup and go from there.

Just for preflop you are looking at quite a few such as Button Raise, BB call, BB fold, 3 Bet, 3 bet fold, 4 bet, 4 bet fold etc. Those will get you started. I have found the folding percentages to be the most helpful. Like if the guy is folding to high percentage of 3 bets then you should be 3 betting very wide until he adapts. Some do and some don't.

Its also important that you know your own bots stats as well as the villains. Sometimes its not so evident and you might have a major leak and not even know it.

well, In in the button of clustering you will you use:
1) aggression:
--pfr
--cbet(every street)
--pfr/vpip
--3bet
--fold to SB
2) hand range:
--vpip
--prf
--call to cbet(every steet)
--call to SB
well even if you develop, based on this, such a model. This is cool, cause it ll give you 4 opponent models. Like aggresive/tight|passive/loose and so on. Its really cool BUT their is HUGE butt is the end. To be shure that modeling is correct you will need hands. For HU, example, to know vpip you will need 100 hands. If opponent is aggresive 3-bet will be around 15%, so you will need at least 300 hands. ETC. Their is the question, how would you know that modeling is correct. So my advise, don't look most used stats. Take a look at stats that is going to "real" value the fastest, such as vpip/pfr, 3-bet(for HU), Flop-Cbet, Fold to SB, and so on.