SUNSETTER


            ^^                   @@@@@@@@@
       ^^       ^^            @@@@@@@@@@@@@@@
                            @@@@@@@@@@@@@@@@@@              ^^
                           @@@@@@@@@@@@@@@@@@@@
                           @@@@@@@@@@@@@@@@@@@@
 ~         ~~   ~  ~       ~~~~~~~~~~~~~~~~~~~~ ~       ~~     ~~ ~
   ~      ~~      ~~ ~~ ~~  ~~~~~~~~~~~~~ ~~~~  ~     ~~~    ~ ~~~  ~ ~~
   ~  ~~     ~         ~      ~~~~~~  ~~ ~~~       ~~ ~ ~~  ~~ ~

[Home] [Download] [Results] [Links]

History of the Crazyhouse Version:

• Lockup when pondering reached max depth(usually a simple forced win) fixed
  
• Support for 64bit linux added
  
• xboard memory command supported, so the hash table size can be set from the xboard menu
  
• xboard analysis command supported, probably broke the support for the bpgn viewer analysis mode?
  
• Fixed all warnings from gcc -Wall, and from valgrind(memory error checker), and all serious warnings from cppcheck. Should be a fair bit more stable now.
  

• Changelog for version 7f seems to have been lost.
  

• minor Eval tuning
• some bugs in Bughouse mode fixed, seems to work now

• KingSafety tuning
• (bug in eval output)

• Razoring tuning
• Speed tuning (unmake copies attackBoards)
• Forcing extension reimplemented, extensions tuned
• Deleted end of variation display
• (learn bug fixed, more #define, faster findFirst, inlining)

• Rewrote king safety eval(), not very finetuned yet.
• (ch. display on mate, linux __forceinline )

• Fixed a nasty pondering bug introduced in 5i.
• Commented out some part of eval() which isnt working properly, will try to fix that next.
• (stdout,display on mate)

• Important! The format for storing data in learn and hash table has changed, so dont use old learn files. You need to have learn in an ini file or as a argument to make it write a learn file ! See [Configuration].
• No more makeHashOld() between moves. Different Hash replacement. This means you can now use as much hash as possible on your system and it should still not lose time because of it in 1 0 games.
• Heaviliy improved pondering. I would be curious how well a sunsetter which has depth set to 3 or so would do by just thinking on its opponent time. Any timecontrol would be possible and the human opponent could decide on the pace of the game ..

• Done some _inline optimization
• 5h seems stable
• Analysis mode implemented for use with BpgnViewer
• Small changes and different IO code for linux by Angrim. Now includes makefile and should compile on linux as well.

• Different material factor in Board control (see eval.cpp)
• Fixed crash in bughouse mode

• Improved Learning, learns now from every game, even close ones. This should completely avoid repeated games and will make this version untestable against earlier ones because it will simply outlearn them.
• Swindle mode, do not sack when I see I get mated. Note that Sometimes when the mate is still too far away I don't see that I get mated.
• Draw thereshold now depending on opps rating
• New pretty aggressive time management. Should play opening faster as soon as it "knows" positions well, and use up to 4x search depth more time when in big trouble. Hope this doesnt make it flag sometimes. We have to watch out for that.

• Warning! A half move now is counted as "4" internally, so you might have to change your settings files.
• Move ordering sorted by To-Squares in 2 steps. Meaning I generate the knight moves to squares where knight moves to used to be good first, and therefore look at them first.
• Some razoring when the last move didn't check, capture, move out of an attack or attack something. This is missing blocked attacks.
• Some eval tuning.

• Fixed some search problems. Completely hopeless positions used to be evaluated as "mate" which resulted in some wrong mate announcements.
• Search changed back to standart alpha-beta. Deleted some very agressive pruning based on a shallower search being at least == alpha. Deleted Futility Pruning. In Crazyhouse the evaluation siwngs between 1 move can just be too high.
• Implemented result learning. When a game is won and we liked the position we got out of the opening for every of our moves in the opening a bonus is stored. The next time we get into this position to the evaluation of this move the bonus is added. The learn file is called learn.bin ; it is created if it doesn't exist. Thanks to Slater for making me do it.
• Improved timing code should prevent "a1a1" crash (thanks to Rodolfo Dall'Armi for reporting).
• Forcing extensions disabled.
• Move generation split into even more parts (see search output).
• Changed the way the captures are ordered and pruned.
• More Information in the tree output (#define gametree)
• Simplified Evaluation, which is the next thing to work on.

• Fixed lots of bugs ( pawn on b2 checks king on a1 now detected :)
• Better Hashing
• Improved Time Management
• Completely new static evaluation
• New recapture extension
• Implemented Null Move
• Implemented Extended Futility Pruning
• Changed the move ordering from History+Killers to oredering by shallow search
• Plausible Move Generation (mainly mate search) near the leaves
• Some MoveGen and Bitboard speed improvements
• Ghost Pieces in Bughouse
• Easy way to change playing style via ini file or commands
• Can annotate a crazyhouse game in bpgn
• Can output a whole search tree in html
• Deleted some usefull stuff, its MSVC source now ;(
• Lots of other things I forgot
• Added some bugs

History of the Chess Version: