Poker-AI.org

Hello,

I am rewriting a few routines on my HUNL bot and could use some help on my river logic. I recently converted to weighted ranges and would like to have somebody look at this logic as I feel I am missing something. Lets assume the villain just bet the river and I am faced with a decision of making a call.

So my bot does a good job of range reduction and for this situation I am including all of his possible value hands and bluffs in his range and excluding any marginal holdings (although he may on occasion turn a pair into a bluff but there is really no real way to know when I guess).

So the villain would never bet 100% of his bluffs so we have to figure out how often he would have to bet them in order to make me indifferent to calling I think. So the formula for that is:

MaxBluff =(BetToCall/(PotSize+BetToCall))

Lets assume its 30%. So to get equity I believe I would weight his bluffing hands at 30% and value hands at 100% correct?

So if so I would be calling with any equity > 50% obviously. Does anybody know at which point I would be value raising back? Is there a formula for figuring that out? I am going to assume its in the 70%+ area but that is just a guess that I have been using.

Any help would be appreciated. Thanks

Maybe page 32 of http://poker.cs.ualberta.ca/publication ... gs.msc.pdf will help.

Thanks. I looked at that and it confirms what I was thinking..but there is a flaw in my thinking. I am assuming everybody is betting optimally and they are not bluffing more then they should be.

I think I should assume that they are bluffing correctly by default and adjust weights based on what river card falls. An example would be betting a bluff when the flush draw or scare card hits. I think they they would be more likely to bluff these types of rivers vs a blank river. Opinions?

Do you believe you know the chance of villain holding any particular hole cards before betting starts on the river?

Yes..I have a rather involved system of ranging a player from pre to the river so I have a complete combinatoric picture of what hands he would reach the river with based on his actions. I think its around 70% accurate so far. I believe I improved that by quite a bit recently by adding in showdown information to verify ranges but I do not have the data to back that up yet.

I decided to double the bluffing weight when a scare card hits. These include AKQJ overcards, 3 flush and str8 draws that fill to broadways, 76,87,98, and T9

One thing I forgot about in this situation is that the villain would also be more likely to bluff all of the hands that contain a blocker to the str8 fill that he reaches the river with as well.

I may be able to help. I can simulate some hands for you and change the hero call %, based on the board and the strength of hero's hand. I can also change the bluffing propencity of the villian. Let me know if you think this would be useful. I have 213 hole/board combinations for the River, here are the first 14.

I finally figured out what to do on this river problem (actually any post flop situation). The solution was to use perceived range stack off equity. So basically you keep track of your perceived range and run equity against his range that he could potentially stack off against you based on his SPR. This solved all kinds of issues all at once. The great thing about this is that it auto adjusts for hands that you might not have accounted for such as slow played hands. It also lets you know if you should even bet at all and just check which is of course extremely helpful when your OOP.

I think I have come to the same conclusion but I'm not sure because I don't understand your jargon.
- http://www.poker-ai.org/archive/www.pok ... f=3&t=4783
- Are you saying that you figure out what villain thinks you hold?
- And then you use that to determine a chance of winning at showdown?
- How do you decide when to bluff and slowplay?
- Do you use this technique just in a single hand or over many hands to figure out when to bluff and slowplay?

Yes..you figure out what you think the villains perception of your range is. Then you determine what hands he can potentially stack off with or hands with X% minimum equity (aka Stack Off Equity). This is based on SPR. Then you run equity with your actual holding vs his stack off range. I only use this before deciding to bet or if I get raised. So if I cant bet and call off his raise then I am forced to check call or check back. You do not want to get blown off your equity so its important to determine this before betting.

I do not use this for bluffing as I am going to be forced to bluff on certain board types/blockers anyway regardless of his range. So what I mean here is if I do not have any showdown equity then I am going to be forced to bluff.

As far as slowplaying goes I think that is mostly dependent on protecting your calling range (for Holdem) or else its rarely the right play. I am working on PLO right now and slowplaying is a bad idea because of all of the draws. You might as well build a pot while you can.

- The best time to bluff is when hero has a weak hand but villain thinks hero has a strong one. With this analysis you know when that happens.
- The best time to slowplay (assuming there is one) is when villain thinks he has a strong hand, but hero has an even stronger one.

You are right. I guess the right time to slowplay would be anytime the stack off equity is >= 55%. Technically it could be 50% but that is probably not strong enough. I think this would have to be randomized somewhere in the 20% area as you wouldn't want to do it all the time however. What do you think?

As far as bluffing I have already incorporated perceived ranges into the equation for holdem:

http://poker-ai.org/phpbb/viewtopic.php?f=24&t=2566

For PLO it is a bit more complicated as you would rarely be bluffing air and if you are it must be incorporated with a number of other factors to "sell it".

- What is stack off equity?
- You have to have a monster hand to slowplay. I can't give you numbers because this is just a theoretical exercise for me at present.
- Yes, you definitely have to randomise slowplay
- I think you should be able to derive a formula for the slowplay frequency using similar principles to the bluffing frequency - from what I remember it generally turns out smaller than bluffing frequency

This is what I was explaining above in which its the perceived range equity vs his potential raising or shoving range. So in this situation villain would definitely have a good hand but hero would have a better hand most of the time.



Yeah its smaller then the bluffing frequency. I think its in the 10-20% area.