search

Zeta36 / chess-alpha-zero

Chess reinforcement learning by AlphaGo Zero methods.
MIT License
2.13k stars 479 forks source link

policy_out dimension #93

Open Vovak1919 opened 4 years ago

Vovak1919 commented 4 years ago

Tell me, how the dimension size was determined in the Dense output layer (None, 1968) in the policy_out header?

Akababa commented 4 years ago

That's the number of combinations of (starting square, ending square) for legal chess moves (in a rank, file, diagonal, or knight), plus some extra indicators for promotions. See the original DeepMind paper for the detailed explanation.