*The painter ChatGPT “A massive, impenetrable fortress built on a rocky cliff, surrounded by towering stone walls with iron gates.”
NIT NPC model
Although we implemented a Texas Hold’em NPC player based on the expected utility function, the behavior is clearly different from that of a human player, and we should implement it in a way that makes good use of bluffing. I have described the characteristics of several player types, but the player type that seems the easiest to implement is the NIT. He is an extremely tight player, but at the same time, his playing style can be described as passive in the sense that he is very timid. Since it hardly uses bluffing, it is relatively easy to implement by program. The purpose of this article is to consider the algorithm and at the same time try to implement it. However, if all the players at the table are NITs, the game may not even be possible if the number of players is too small, so I think I’ll try implementing the game as something like Reader vs. 8 NITs. Who can survive? (from Mobile Suit Gundam)
NIT Betting Strategy
I speculate that NIT’s preflop behavior pattern will be as follows:
- Hand ranges are AA, KK, QQ, AKs, and AKo, and anything else is folded. No mixed strategies are used.
- If they are the first raiser, they choose 3x raise. If they are the second raiser, they should choose 3-bet and 3x raise. If they are the first raiser and the bet is over the 3-bet range, they choose to call.
- Once they enter, they generally do not fold even if other players raise, but choose to call any raises above 3-bet.
If implemented like this, the preflop strategy for NIT should be fine. The strategies for the flop and turn are as follows:
- On the flop, if they cannot get a pair or better (only with AK), they choose to check or fold (fold if they are raised). The same goes for the turn.
- If the community cards contain a straight draw (open-ended, gutshot) that does not result in a straight for them, or a backdoor flush draw that does not result in a flush or flush draw for them, they can choose to check or fold.
- If the situation does not fall under 1.2., bet CB for 50% of the pot (no donk bets). If another player raises, choose to call.
The strategy on the river is as follows:
- If they do not have a hand of one pair or better, check or fold.
- If the community cards contain a straight draw (open-ended, gutshot) that does not result in a straight for them, or a backdoor flush draw that does not result in a flush or flush draw for them, they can choose to check or fold.
- If the situation does not fall under 1.2., bet CB for 100% of the pot. If another player raises, choose to call.
Improvement
Well, surely that would mean that if they raise you preflop you just have to fold, and then you can steal the blinds as much as you want, right? So let’s think of some ways to improve this.
- For middle pairs (JJ, TT, 99, 88, 77) and high ranks (AQo, AQs, AJs, ATs, KQs), there is a certain probability (50%) of making a bet in the same way as for AA, KK, etc.
- In positions with blinds, such as SB or BB, bets are made on hands with a winning probability discounted by the amount of the blinds in the same way as for AA, KK, etc. Make it a guaranteed raise with middle pairs (JJ, TT, 99, 88, 77) and high rankings (AQo, AQs, AJs, ATs, KQs).
- Even if they don’t have a pair or hand on the flop or turn, if they have an open-ended or flush draw, they will bet the CB as if they had a pair (performing a semi-bluff). They generally do not bluff on the river (if they slip up, they simply check or fold).
The rules are set and the player does not predict the other players’ hands, but once they get to this stage, it may become difficult to defeat them. It would be good if the relatively strong tight passive player feel could be expressed well… It’s not good to have their strategy fixed on the river, but since their hand is strong to begin with, it seems difficult to polarize. If it were to be implemented, it would probably have a 1/4 chance of going all-in (pure bluffing) even if it was a high card that was a straight draw or flush draw that fails. However, if we implement it to this extent, it may no longer feel like NIT. For now, I have implemented the original NIT and the improved version of NIT, so why not try to defeat it? (Although you may not be able to win)
Comments