1. zurichess
  2. zurichess
  3. zurichess
  4. Source

CHANGELOG.md

Source Diff History

zurichess / CHANGELOG.md

Full commit
Annotate Embed Raw

History

Versions are named after Swiss Cantons in alphabetical order.

zurichess graubuenden (stable)

16.Aug.2016

The theme of this release is tweaking and improving the static evaluation.

An overview of the most important changes is:

  • Use a new version of the Go compiler for increased search speed.
  • Hash quiescence search and static evaluation.
  • Improve Null Move Pruning:
    • Consider NullMove irreversible when testing draws.
    • Prune only if static evaluation is at least equal to the upper bound.
    • Prune at all depths.
  • Improve evaluation:
    • Evaluate king near own passed pawns.
    • Evaluate knight's and bishop's position (psqt).
    • Evaluate backward pawns.
  • Simplify Late Move Pruning and History Leaf Pruning.
  • Check extend all moves with positive SEE.
  • Use shell sort instead insertion sort for move ordering.
  • Tweak the time control to avoid forfeits for long games.
  • Usual code clean ups, speed ups and bug fixes.

A few features have been introduced:

  • Skill Levels: Zurichess supports 11 skill levels of playing. The default skill level, 0, produces the strongest play, and should be used for rating lists. At skill level 10 zurichess plays about 700 Elo weaker, and it is recommended for more casual players. To change the skill level input: setoption name Skill Level value 7
  • Multi PV: zurichess has the ability to print multiple principal variations. To set the desired number of principal variations input: setoption name MultiPV value 3
  • Theban chess: zurichess supports a chess variant popularized by Kai Laskos which is played with standard chess rules starting from the unorthodox position 1p6/2p3kn/3p2pp/4pppp/5ppp/8/PPPPPPPP/PPPPPPKN w - - 0 1. No playing strength tests or improvements were done other than fixing the crash of the previous version.

zurichess glarus

17.Apr.2016

The theme of this release is king safety and leaf pruning. ELO is about 2554 on CCRL 40/40 and 2402 on CEGT 40/4.

  • Improve futility conditions. Geneva's futility is a bit too aggressive and causes lots of tactical mistakes.
  • Add History Leaf Pruning similar to https://chessprogramming.wikispaces.com/History+Leaf+Pruning.
  • Improve pawn evaluation caching. Also cache shelter evaluation.
  • Improve king safety using number of simultaneous attackers.
  • Improve time control. Timeouts should be extremely rare now.
  • Small tunining of LMR and NMP conditions.
  • Micro-optimize the code for the future Go compiler. Next version will see big speed up.
  • Move Position.SANToMove to https://bitbucket.org/zurichess/notation
  • Move Polyglot hashing to https://bitbucket.org/zurichess/hashing
  • Usual code clean ups, speed ups and bug fixes.

zurichess geneva

04.Dec.2015

The theme of this release is improving evaluation. ELO is about 2475 on CCRL 40/40 and 2320 on CEGT 40/4.

  • Implement fifty-move draw rule. Add HasLegalMoves and InsufficientMaterial methods.
  • Improve move ordering: add killer phase; remove sorting.
  • Improve time control: add more time when the move is predicted.
  • Add basic futility pruning.
  • Switch tuning to using TensorFlow framework. txt is now deprecated.
  • Evaluate rooks on open and half-open files.
  • Improve mobility calculation.
  • Tweak null-move conditions: allow double null-moves.
  • Usual code clean ups, speed ups and bug fixes.

zurichess fribourg

04.Sep.2015

The theme of this release is tuning the evaluation, search and move generation. ELO is about 2442 on CCRL 40/40.

  • Move to the new page http://bitbucket.org/zurichess/zurichess.
  • Evaluate passed, connected and isolated pawns. Tuning was done using Texel's tuning method implemented by txt.
  • Add Static Exchange Evaluation (SEE).
  • Ignore bad captures (SEE < 0) in quiescence search.
  • Late move reduce (LMR) of all quiet non-critical moves. Aggressively reduce bad quiet (SEE < 0) moves at higher depths.
  • Adjust LMR conditions. Reduce more at high depths (near root) and high move count.
  • Increase number of killers to 4. Helps with more aggressive LMR.
  • Improve move generation speed. Add phased move generation: hash, captures, and quiets. Phased move generation allows the engine to skip generation or sorting of the moves in many cases.
  • Implement setoption Clear Hash.
  • Implement pondering. Should give some ELO boost for online competitions.
  • Improve move generation order. Picked the best among 20 random orders.
  • Prune two-fold repetitions at non-root nodes. This pruning cuts huge parts of the search tree without affecting search quality. >30ELO improvement in self play.
  • Small time control adjustment. Still too little time used in the mid game. Abort search if it takes much more time than alloted.
  • Usual code clean ups, speed ups and bug fixes.

zurichess bern

25.Jun.2015

This release's theme is pruning the search. ELO is about 2234 on CCRL 40/4.

  • Implement Principal Variation Search (PVS).
  • Reduce late quiet moves (LMR).
  • Optimize move ordering. Penalize moves threatened by pawns in quiescence search.
  • Optimize check extension. Do not extend many bad checks.
  • Change Zobrist key to be equal to polyglot key. No book support, but better hashing.
  • Add some integration tests such as mate in one and mate in two.
  • Usual code clean ups, speed ups and bug fixes.

zurichess basel

28.Apr.2015

This release's theme is improving evaluation function.

  • Speed up move ordering considerably.
  • Implement null move pruning.
  • Clean up and micro optimize the code.
  • Tune check extensions and move ordering.
  • Award mobility and add new piece square tables.
  • Handle three fold repetition.
  • Cache pawn structure evaluation.
  • Fix transposition table bug causing a search explosion around mates.
  • Prune based on mate score.

zurichess appenzeller

23.Feb.2015

This release's theme is improving search. ELO is about 1823 on CCRL 40/4.

  • Clean code and improved documentation.
  • Implement aspiration window search with gradual widening.
  • Improve replacement strategy in transposition table.
  • Double the number of entries in the transposition table.
  • Develop zuritest, testing infrastructure for zurichess.
  • Fail-softly in more situations.
  • Implement UCI commands go movetime and stop.
  • Add a separate table for principal variation.
  • Add killer heuristic to improve move ordering.
  • Extend search when current position is in check.
  • Improve time-control. In particular zurichess uses more time when there are fewer pieces on the board.

zurichess aargau

22.Jan.2015

This is the first public release. ELO is about 1727 on CCRL 40/4.

  • Core search function is a mini-max with alpha-beta pruning on top of a negamax framework.
  • Sliding attacks are implemented using fancy magic bitboards.
  • Search is sped up with transposition table with Zobrist hashing.
  • Move ordering inside alpha-beta is done using table move & Most Valuable Victim / Least Valuable Victim.
  • Quiescence search is used to reduce search instability and horizon effect.
  • Simplified evaluation function with tapered eval.