Poker-AI.org

Hello everyone,

so i read the papers on Pluribus and Libratus, but did not find any information on what programming language the authors use to run the CFR simulations.
As their algorithm runs for several days on a supercomputer i imagine it has to be very optimized. So what is currently the proper language to implement high-scale CFR calculations. Also is it correct that both pluribus & libratus don't use any neural nets as it is used e.g. for DeepStack?

Some answers on this would be much appreciated, thank you.

I know that the university of Alberta who made Deepstack and some other famous bots before that use C. Personally I use C++, there are many C and C++ libraries that you can use and it's generally one of the fastest languages, arguably even the fastest.

I use Scala and Java. You could probably make it faster using C or C++ but I find the time saved in writing Scala a significant factor.
viewtopic.php?f=24&t=3235 suggests C/C++ is 2 to 3 times faster than Java which seems plausible to me. I imagine C# is also an option.

Neither Pluribus nor Libratus used neural networks, but Noam Brown has used neural networks in his research https://www.cs.cmu.edu/~noamb/research.html

After the language there are also frameworks etc. to consider. You will probably need some inter process communication framework if you want to go really big, Alberta used MPI for Cepheus, the bot that solved FL. I don't remember which MPI implementation it was only that it was NOT open mpi.

C/C++ for speed, Java for convenience.

If I had to write something with the speed of C/C++ I'd take a good look at Rust these days.