Poker-AI.org

Hi guys. Part of the algorithm I'm developing requires to calculate various measures associated with the handstrenght for each single hand possible given the board. Ideally, I would like to tackle the problem with a monte-carlo simulation. I'm planning to code a wrapper in cython for poker-eval, becouse i'm less familiar in C++ than Python and i need to interface it to the rest of my app in Tensorflow python.
Any advice? This should be a good starting point https://pypi.python.org/pypi/eval7/0.1.2 .

I'm an amateur at C++ and poker-source really made my head hurt when I looked at it. I reckon the twoplustwo evaluator is faster and using it is dead simple. (assuming you don't have memory limitations and you are looking for 7 card holdem evaluator)

I use this calculator: https://github.com/ktseng/holdem_calc/b ... /README.md
Its a Monte Carlo simulation for python, with some minor adjustments you can make work in python 3+. Take in consideration that on if you give a board, it calculate the exact probability even if you specify you don't want the exact calculation. If you want to run a simulation you have to delete "if given_board is not None" from run_simulation() in holdem_calc.py

Thanks spears and ProfundMadman for your suggestions. The handstrength simulator is the last part that I do not understand well how to do it of my project, for this I would like to solve it fairly quickly though not in the best way. I have difficulty working with a c ++ project, while one in pure python is too slow.


In conclusion, I think I have to rewrite my poker hand evaluator directly in tensorflow. I'm working on google cloud, so the ram and cpu limits are given by those that I can reasonably afford. What do you think is the most appropriate algorithm?
I'm thinking this:

- Batch, build the kevin's rule rank hash table for all 7 cards hands ( 133784560 hand ranks, every saved as a 4 byte integer so only 0,5 GB, maybe i've made a mistake seems too small). 64 cpu may cost 0.10-0.20$/hour, tips for parallelize this process?

- Online, for every hand simulate for N times a random opponent hand and a random board and update the win total of the hand through a table lookup. Here I can parallelize over the simulation (or over the hands).

-Return the win totals array / N.

I plan to make the batch script building the hash table in cython (reusing the code of the project in my first post) and then to use it in a pure tensorflow python API app.