lifrordi / DeepStack-Leduc

Example implementation of the DeepStack algorithm for no-limit Leduc poker
https://www.deepstack.ai/
878 stars 211 forks source link

How to speed up solving from the turn card is dealt #30

Closed zhoujz10 closed 6 years ago

zhoujz10 commented 6 years ago

Hi @lifrordi Thank you for releasing the code of DeepStack-Leduc.

In your paper, you mentioned that you generated ten million situations for the turn network in 175 core years, which means 551 core seconds for each situation.

I tried this but my code is much slower and it needs thousands of core seconds to generate one situation. There is too much matrix calculation.

So, would you please share with me how to optimize this process? Did you use some abstraction?

Thank you very much!

ghost commented 6 years ago

Try out the CFR+ implementation written in C with optimized code and cluster support: http://poker.cs.ualberta.ca/cfr_plus.html

zhoujz10 commented 6 years ago

@Sanjo Thank you very much for sharing this with me, I'm reading the codes now.