Changes in Arasan 23.2:
1) Fixes to singular extension code.
2) Change static null pruning margins.
3) Fix compile errs in tracing code.
4) Don't reset debug prefix when executing "new" command.
5) Fix myname feature (for CECP mode).
6) Package correct network file for Windows.
7) Build fixes for Cygwin
Hi there, version 32 is out ! It took quite some time to get there (mostly because improving Stash gets harder as time passes, but also because I have been quite busy with my studies lately), and I'm really happy with the result.
Improvements are about 65 Elo in self-play, probably a bit lower against other engines.
This release adds approximately 100 elo in self play. This increase in strength can be attributed to both further refinements to the search function and an improved network in roughly equal measure. As with the previous release, the 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. Associated training code can be found at the "seer-training" repository: https://github.com/connormcmonigle/seer-training/tree/shared-affine.
Notably, this release adds support for both ponder and Syzygy EGTB probing.
Another release with an improved NNUE evaluation and small search patches (same as last time).
The newest network was trained using a custom trainer on 1.2B FENs from Berserk 6 self play games. Most of the improvements for this release come from tweaks in the trainer and not a largely improve architecture or better data. I was hoping to improve the architecture for this release, but all attempts at this time have failed (miserably I may add).
Minor search patches are included in this release as well.
New features included in this version are evaluation terms for king safety and pawn structure. Both are still quite basic and have much lacking, but I believe them to be solid designs to build on in future versions. Addtionally, all evaluation terms have been retuned on a larger dataset, yielding some minior gains. However, most of the strength gain in this new version comes from slight tweaks to the pruning parameters and conditions for late-move reductions, null-move pruning, and futility pruning. Lastly, there are some non-strength gaining tweaks here and there, code clean-ups, and the command line interface is much more user friendly.
Black Marlin 2.0 is composed of various improvements in search, a new neural network trained for a longer amount of time and a completely new time management system.
Blue Marlin is focused on king attack and sacrifices. It provides very sharp and attractive games.
This time only Blue Marlin 14.6 is released. For reasons unknown to me, the strongest beta candidate of Swordfish 14.6 failed in 2 final tournaments.
What's strange, more aggressive engine Blue Marlin 14.6 passed easily all tests. The engine is very tactical and suspiciously strong.
@Superbatil wrote:Sir Archimedes can I suggest of Eman chess engine on android? Thanks in advanced. Advanced Merry Christmas.