Poker-AI.org

Hi,
My bot is rule based and coded in oh-script of the OpenHoldem.

What shall I use (if its even possible) to get the bot ranges in real time in any given spot, in order to use it in decision making?
I've currently "mapped" the strategy offline using CardRunnersEV and implement those percentages manually (ie bluff %, call % etc).

I dont understand the question "the bots own ranges". Do you mean the bots perceived range to the villain? If so you need to figure out what the villains HUD would see. There is a thread here based on how I figured it out which is not easy. For the bot to sell a bluff effectively it will have to do this.

Using CardRunnersEV is an interesting way of doing it but I am guessing it will not adapt enough so you will have to modify the baseline strategy quite a bit in order to take advantage of villains tenancies. Other then that its probably a good way to start.

No, I was talking for about the exact % of different parts of bot's ranges.

Example: on K72 6 8 vs. some specific line, bot may hold 30% 7x, 45% Kx+, 25% air.
I want to be able to use this percentages in decision making. Right now im only "guessing" them by study offline with CREV.


I was thinking on something based on montecarlo simulations...?!?

oh ok I see what you mean (I think). What I had to do was run simulations on certain lines to see what the bots range consisted of to ensure it was balanced. An example is I ran sims on the bot doing a preflop raise and c betting every flop. By doing this I noticed it was not c betting enough air as it was value heavy. So I adjusted it until it was correct. It gave me a wealth of information and you can run sims on every aspect of the bots behavior. I was mostly interested in the bots folding tendencies which good players will exploit quickly

Its hard to tell what is going on with the bots ranges until you have a ton of data.

If your bot is rule-based you should be able to calculate those values exactly.
You'd have to create a game tree, walk it and match it to your scenarios and then do the math.

Or you can run simulations and approximate it that way. Either way, a tree representation will be the easiest to generate and store those numbers.

_________________
Cheers.