|
|
2008: The first WinBoard version. The engine is pretty much as it was in 1999, except now it is a 32-bit program instead of a 16-bit one. A few additions to the position evaluation code: pawn balance, square of the pawn, passed pawns, backward pawns... Plus some outlandish concepts such as piece mobility (!!!). Estimated playing strength: 2,130 Elo points (CCRL).
2009, 2010: A separate function to handle quiescence search. Better move ordering with internal iterative deepening. WAXMAN's rating is up to 2,310.
2011, 2013: First experiments with selective searching using late-move reductions (LMR). Static exchange evaluation (SEE) in quiescence search to weed out losing captures without actually searching them. No Elo gain at regular time limits, but WAXMAN's blitz rating is up!
2014: First successful attempt at selective searching. Much more aggressive LMR backed up by a verification search. WAXMAN's rating reaches 2,390 -- sadly, a high-water mark for some time to come.
2015, 2016, 2017: Experiments with different search refinements. Learned a lot but nothing to show for it in terms of playing strength. Small improvements elsewhere: better time management in games with conventional time limits, tapered evaluation, support for WinBoard hint and analysis mode.
2020: Late-move pruning (LMP) in addition to LMR. No more history heuristic -- simpler code, no ill efects. Rating up by 50 points. Also new: a human-readable display of the principal variation.
2022: The work on a new "minimalist" evaluation function is taking longer than expected, so I'm sticking with the old one for now. In the meantime, some changes to the search code: (pseudo-)random search extensions, safer futility margins.
| ©1990-2022 Ivan Bacigal |