https://pixeldrain.com/u/iVyNRAn3
Stockfish 220419 with embedded network:
https://pixeldrain.com/u/jY8aWbQP
Stockfish 220419 (OEX):
https://pixeldrain.com/u/t2oxeRKc
This release introduces an evaluation function based on a NNUE-style architecture of input size 728, with 8 buckets of 256 hidden neurons which is chosen depending on the number of pieces left on the board. This network was trained on 125 million positions from self-play games using Weiawaga v4.0. The only new search change is the introduction of reverse futility pruning. The release also includes many minor optimizations, and is ~350 Elo stronger than Weiawaga v4.0.
This release adds roughly 100 Elo in self play. The majority of this strength improvement can be attributed to a variety of miscellaneous optimizations and an improved network trained by way of several further self play training iterations. Additionally, time management was revisited with this release which should notably improve performance at shorter time controls. As with the prior release, the embedded network was trained solely on data originating from Seer produced by way of several self play iterations starting with the initial "retrograde learning" network bundled with v2.0.0-v2.1.0 as a base.
apologia wrote:
Aligned with stockfish patch
Negative extension for ttMove that is less than alpha and value in the context of singular extensions
Improved Shashin theory
Aligned with stockfish patch
Negative extension for ttMove that is less than alpha and value in the context of singular extensions
Archimedes wrote:Improved Shashin theory
Aligned with stockfish patch
Negative extension for ttMove that is less than alpha and value in the context of singular extensions
https://github.com/amchess/ShashChess/releases/tag/22
ShashChess 22:
https://pixeldrain.com/u/GS3EPHi9
In the case, this version do not work, try the following one.
ShashChess 22, compiled with Android NDK:
https://pixeldrain.com/u/qnAM9FQp
|
|