I released Donna v2.0 this morning -- watching Graham's gauntlets live on TLCV inspired me to make it happen.
In the new version I revamped transposition table, refactored quiescence search, tuned up some evaluation parameters, and fixed several nasty bugs discovered by analyzing CCRL games. It's hard to predict ELO score but in self-play Donna 2.0 outplays previous version 78 to 32. For curious minds compete lists of changes is available at https://github.com/michaeldv/donna/compare/v1.0...v2.0
Few things I've learned while building Donna:
- Your chess engine hates you when you are working on a new version
- Your chess engine makes weak moves on purpose
- Your chess engine will find a way to make a stupid move
To build Donna you will need Go v1.3 or later installed on your system (see http://golang.org/dl/). From Donna source directory run "make" or "go build"
command:
I haven't tested it though so view it as "unofficial"
-------------------------------------------------------------------------------------
Ok Mike thanks........look forward to testing it in some upcoming tournaments
Thanks for testing, seems like a nice boost. Wonder how it's going to stack up at 40/40.
Donna 1.0 had few major flows: buggy and inefficient implementation of transposition table and pawn cache, nasty KPK evaluation bug, and somewhat insane time allocation where it could spend up to 25-30% of remaining time on single move. All despite 300+ unit tests -- almost confirms my theory that passing test suite proves that there and then the test suite passes
