Releases: crybot/Napoleon
Napoleon 1.8
- Implemented a form of static null move pruning;
- Added PVS search at root nodes;
- Improved late move reduction;
- Implemented Pondering;
- Implemented basic passed pawn and king security evaluation;
- Other small improvements and refactoring;
Results:
Test results show an improvement of about 130 elo with pondering disabled.
The new testing framework requires 2400 games at TC 15"+0.3" for each functional change.
Includes statically linked PGO optimized windows binaries and Fauzi's opening book used for testing.
NOTES: Current version only works with bmi2-compliant Intel cpus, which means at least Haswell. I'm working to provide cross platform/architecture binaries.
Napoleon 1.7
- Implemented a slightly modified form of Lazy SMP;
- Safe concurrent access to Transposition table is ensured by very fine lock granularity:
- a C++11 implementation of a spinlock was used to lock each bucket of the TT;
- Added UCI 'Threads' option to set the number of concurrent threads to search with;
Results:
Parallel search does scale quite effectively, reaching depths 4 times faster;
300 games with TC 3'/0.5 against the old version showed an improvement of about 130-140 elo on an i7 quad core machine.
Includes statically linked PGO optimized windows binaries and Fauzi's opening book used for testing.
Napoleon v1.6.0 beta
beta pre-release
only for testing purpose
Napoleon v1.5.0
-Introduced a form of enhanced razoring
-Implemented isolated pawns penalization
-Added game phase-dependent bonuses for pieces (Rook, Knight and Pawn)
-Added x64 PGO optimized executable
Results:
6000 bullets self games at 40/20 (40 moves for 20 seconds)
show an improvement over the last version of about 30 elo.
Napoleon v1.4.0
-New evaluation function
-New mobility bonus calculation
-Added bishop pair bonus
-Exclude squares controlled by enemy pawns in knight mobility
-Bishop and knight have now the same value (330)
-Implemented tapered evaluation
-Added support for selective depth UCI pv-info
-Generalized Piece-Square-Tables update
-Code refactoring
Results:
Self play games seem to show a gain of 100-150 points of elo.
Tests are still running.