Expositor is a UCI-conforming chess engine for AMD64 / Intel 64 systems.
Author: Kade Rating CEDR 15'+10"=3048, Rating CEDR 3'+3"=2971
Releases
If you know the microarchitecture of your processor, try using a binary from the appropriate specific/ directory of a release archive.34 If you don't know the microarchitecture of your processor but you do know which features it supports, try using a binary from the appropriate generic/ directory of a release archive. See the file named "extensions" in this repository for more information.
The binaries include the default network, so you do not need to download a separate copy and do not need to set the EvalFile UCI option.
If you know the microarchitecture of your processor, try using a binary from the appropriate specific/ directory of a release archive.34 If you don't know the microarchitecture of your processor but you do know which features it supports, try using a binary from the appropriate generic/ directory of a release archive. See the file named "extensions" in this repository for more information.
The binaries include the default network, so you do not need to download a separate copy and do not need to set the EvalFile UCI option.
Building
If you'd like to compile Expositor from source, you will need to use a recent nightly toolchain.
To build Expositor on Windows, run the build.bat script. (This sets the VERSION, BUILD, and RUSTFLAGS environment variables and then invokes cargo build --release.)
To build Expositor on Linux, run the build script. (This needs to be done from within the repository, since it uses Git to automatically determine the version number.)
Issues
If you find any bugs or have any questions, please file an issue on Github or send me a message.
Pending
The to-do list has gotten to be rather long and variegated. My current plan is to tackle the smaller items first (aiming for a release in March or April) and then to begin work on some larger projects:
Tuning Search None of the search constants have been tuned! I expect to be able to wring a fair amount of playing strength out of that.
HCE Bootstrapping The neural network is currently trained from positions scored with Stockfish. I'd like to write an evaluator that replicates the personality of early versions of Expositor, train a network from positions scored by Expositor using that evaluator, then train another network from positions scored using the previous network, and so on.
Experimental Network Architectures I'd like to try using different input features and play around with small convolutional networks.
Error Tracking Search I've been reading and thinking about this since I started chess programming and at one point it was my primary focus. I'd like to pick it up again, deliver a working proof of concept of my ideas, and then write up my findings.
Infrastructure I'd like to write my own automation system (à la OpenBench) that will kick off SPRT testing whenever I push a commit.
These are processed by Expositor with her quiescing search and the leaves from those searches are then scored with Stockfish.
Most of the positions used for perft tests are from the Zahak repository.
Particular thanks to Jeremy Wright for help with main search. Many techniques were implemented from his descriptions and use parameter values he suggested.
Individual statistics: Expositor 2WQ23 - 72 games (+ 31,= 14,- 27), 52.8 % (15'+10")
Jackychess 0.9.14 : 1 (+ 1,= 0,- 0), 100.0 %
Kouri 1.15 : 1 (+ 1,= 0,- 0), 100.0 %
Sapeli 2.0 : 1 (+ 1,= 0,- 0), 100.0 %
Rustic Alpha 3.0.0 : 1 (+ 1,= 0,- 0), 100.0 %
Iathena 2020-12-19 : 1 (+ 1,= 0,- 0), 100.0 %
FracTal 1.0 : 1 (+ 1,= 0,- 0), 100.0 %
Chareth 0.1.0 : 1 (+ 1,= 0,- 0), 100.0 %
Quokka 2.1 : 1 (+ 1,= 0,- 0), 100.0 %
Monochrome r203 : 1 (+ 1,= 0,- 0), 100.0 %
Stockfish 14.1 : 2 (+ 0,= 1,- 1), 25.0 %
Pulse 1.7.3 : 1 (+ 1,= 0,- 0), 100.0 %
SugaR AI 2.50 : 2 (+ 0,= 1,- 1), 25.0 %
Fisherov 0.98i : 2 (+ 0,= 0,- 2), 0.0 %
BLANK 1.3.0 : 1 (+ 1,= 0,- 0), 100.0 %
Lynx 0.9.0 : 1 (+ 1,= 0,- 0), 100.0 %
MisterQueen : 1 (+ 1,= 0,- 0), 100.0 %
Lc0 0.28.2 : 4 (+ 0,= 1,- 3), 12.5 %
Weiawaga 4.0 : 1 (+ 1,= 0,- 0), 100.0 %
Popochin 4.3 : 1 (+ 0,= 1,- 0), 50.0 %
Prophet 4.1 : 1 (+ 1,= 0,- 0), 100.0 %
Anka 0.6.3 : 1 (+ 1,= 0,- 0), 100.0 %
Eman 7.80 : 2 (+ 0,= 0,- 2), 0.0 %
Dragon 2.6.1 : 2 (+ 0,= 1,- 1), 25.0 %
ClassicAra 0.9.9 : 5 (+ 2,= 3,- 0), 70.0 %
SlowChess 2.83 : 2 (+ 0,= 0,- 2), 0.0 %
Leorik 1.0 : 1 (+ 1,= 0,- 0), 100.0 %
Fatalii 0.1.0 : 1 (+ 1,= 0,- 0), 100.0 %
Tantabus 2.0.0 : 1 (+ 1,= 0,- 0), 100.0 %
Koivisto 7.14 Ipman : 2 (+ 0,= 1,- 1), 25.0 %
Journeyman 1.0 : 1 (+ 1,= 0,- 0), 100.0 %
Eggnog 3.0 : 1 (+ 1,= 0,- 0), 100.0 %
Corchess 250222 : 2 (+ 0,= 0,- 2), 0.0 %
Stockfish 010322 : 2 (+ 0,= 1,- 1), 25.0 %
Frozenight 1.0.0 : 2 (+ 2,= 0,- 0), 100.0 %
Rebel 14.2 : 2 (+ 0,= 0,- 2), 0.0 %
Stockfish 250222 Ivec : 2 (+ 0,= 0,- 2), 0.0 %
Dumb 1.9 : 2 (+ 2,= 0,- 0), 100.0 %
BrainLearn 16 : 2 (+ 0,= 1,- 1), 25.0 %
Lozza 2.2 : 1 (+ 1,= 0,- 0), 100.0 %
Little Goliath 3.15.4 : 1 (+ 0,= 1,- 0), 50.0 %
Frozenight 2.0.0 : 1 (+ 1,= 0,- 0), 100.0 %
Princhess 0.7.0 : 1 (+ 1,= 0,- 0), 100.0 %
Walleye 1.6.0 : 1 (+ 1,= 0,- 0), 100.0 %
CorChess 010422 : 2 (+ 0,= 0,- 2), 0.0 %
ProteusSF 006 : 2 (+ 0,= 1,- 1), 25.0 %
Stockfish 010422 Ivec : 2 (+ 0,= 0,- 2), 0.0 %
Kayra 1.3 : 2 (+ 0,= 1,- 1), 25.0 %
Expositor for homepage
Comments
Post a Comment