http://sourceforge.net/p/chesser/
This version is around ~2000 (scores 70-100% against FMax and 50-70% against CuteChess) in lightning controls. It can gain dramatic strength in relatively longer controls like 40/4 (sometimes ~2200!) Although the earlier versions could do scores as high as 50% against Buzz (~2250), but this version doesn't. I've ignored that thing for now.
Here are some new things:
=> Added hardware instruction support (see bitops.cpp).
Perft is now 30% faster, and so is the search. Thanks to Jose Velosca for advising!
And for people who wish to compile it, make sure if you've a 64-bit compiler, you comment out IS32BIT in file bitops.cpp.
3. Some general speedups... like speeding up the heavily used isOtherKingAttacked() etc.
4. Improved evaluation. Cleaned up some code.
5. Slightly Improved makemove.
6. Added Aspirational Windows. But I failed to add LMR (due to the search taking both
ply and depth as argument. depth is not a problem, but what ply is depth-2?).
I've successfully implemented some of my rediscovered ideas (like using Smaller
Window Size when at relatively higher depth).
Just for this, I had to re-understand alpha-beta, and PVS, and Null-Move pruning...
7. Improved Aspirational Windows (thank you Stockfish!).
For example, before 8 plies, WindowSize of 40 is used, and after that, WindowSize of 30 is used. This slightly improves the performance.
I've put my effort in Chesser as much as I could. Although it has only two public versions: v1.9 and this one, there are _thirteen_ private versions, three of them being my failed tries to speed up, and the other ten the normal versions. Every version was a slight improvement. I've been working on Chesser for many months now. I think now it's time to write my own engine. But I must admit I learned a lot from Chesser:
(i) Strength of the engine doesn't matter to me now, as much as clean code does. Strength comes _almost_ automatically with concise and clean code. Chesser is a very good example of unclean code and problems caused by it. So I'm writing clean code as much as I can in Yaka, and in my other projects (i.e Tantray Pi, Tratth Factorial)
(ii) I learned how search _really_ works
. This was the steepest curve of my chess programming experience (with magic bitboards, which I still don't understand enough ). I've heard people saying the same.(iii) Only speed doesn't matter. When I heard people saying Komodo beats Stockfish with 200 KNPS, I was quite astonished. Chesser is also a kind of engine that has a very good evaluation function, mainly due to hand-crafted psq tables. However, Chesser is not *that* good tactically. For instance, a typical Chesser vs. Buzz game goes like this: both sides are equal... Chesser showing +0.30, Buzz showing -0.01... Chesser showing +0.70, Buzz showing -0.11... Chesser a pawn up... Chesser makes decisive blunder and its all over. This happens almost 75% of the time, seriously! So, I won't be that much surprised to see Chesser rated 2200+ in 40/40. Ofc, what ever I'm saying is theoretical...
(iv) Good chess programming needs ideas... Ideas might not necessarily mean something you're gonna publish a paper on...
Sorry for a long post... Hope you have a look at LMR and Aspirational Windows of Chesser.
Happy Programming and Happy Testing!
