lifrordi / DeepStack-Leduc

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

Converts hand range to bucket #25

Closed DWingHKL closed 6 years ago

DWingHKL commented 6 years ago

@lifrordi hello, I was still confused about convert hand range to bucket My question 1 is: According to your reply. Deepstack arrange buckets by the player's expected hand strength during training and during the continuous re-solving process. As we know normal EHS is based on the uniform of opponent private cards. But both in training and continuous re-solving process we can get opponent range(private cards distribution). So Deepstack use normal EHS(uniform) or use the opponent current range to real time compute EHS. If is later one, in my test it need more than 1 minute to compute EHS in real time. How do you speed up that Deepstack run fast within 10s? Question 2 is: normal card abstraction is do offline and store the result in file, and load the card abstraction file when playing and run cfr. So Deepstack do buckets offline or online ? offline compute Histogram and use EMD and run k-means to produce result?

lifrordi commented 6 years ago

The buckets were pre-computed offline as you suggest.