Ajax Chess 2.0

[gladius]

http://forwardcoding.com/projects/ajaxchess/chess.html

The chess engine is quite a bit stronger now. I still highly recommend Chrome, as I get about 130k NPS on my machine, while other browsers are not nearly as good.

It has no pawn evaluations at all, so the good chess players here should be able to exploit that fact (I can't ).

New features:
- Mobility evaluation (thanks Fruit)
- Bishop pair
- Rep-draw detection
- Better null-move pruning (thanks Stockfish)
- Better LMR (and again, thanks Stockfish)
- Bugfix with using hash move
- Some speed optimizations

[gerold]

Thanks Gary.

Best,
Gerold.

[Albert Silver]

http://forwardcoding.com/projects/ajaxchess/chess.html

The chess engine is quite a bit stronger now. I still highly recommend Chrome, as I get about 130k NPS on my machine, while other browsers are not nearly as good.

According to latest tests, Chrome 6 is the king of scripts.

[gladius]

According to latest tests, Chrome 6 is the king of scripts.

It certainly seems so in practice as well as benchmarks. Chrome is significantly faster than Firefox, Opera and IE. I haven't tested Safari though.

[Albert Silver]

According to latest tests, Chrome 6 is the king of scripts.

It certainly seems so in practice as well as benchmarks. Chrome is significantly faster than Firefox, Opera and IE. I haven't tested Safari though.

I played a quick game against it. The game was in Chrome 6 on my laptop and was going at around 145 kns. It produced a strange illegal move after getting mated. I'm sending the game via PM so you can see.

[zullil]

http://forwardcoding.com/projects/ajaxchess/chess.html

The chess engine is quite a bit stronger now. I still highly recommend Chrome, as I get about 130k NPS on my machine, while other browsers are not nearly as good.

It has no pawn evaluations at all, so the good chess players here should be able to exploit that fact (I can't ).

New features:
- Mobility evaluation (thanks Fruit)
- Bishop pair
- Rep-draw detection
- Better null-move pruning (thanks Stockfish)
- Better LMR (and again, thanks Stockfish)
- Bugfix with using hash move
- Some speed optimizations

Thanks. A good opponent for me!

One suggestion: Display the NPS as an integer, without the fractional part.

For example,

Code: Select all

Ply:8 Score:-165 Nodes:167309 NPS:166808 d5 Nc3 dxe4 d3 exd3 Bxd3

instead of

Code: Select all

Ply:8 Score:-165 Nodes:167309 NPS:166808.57427716852 d5 Nc3 dxe4 d3 exd3 Bxd3

[IanO]

According to latest tests, Chrome 6 is the king of scripts.

It certainly seems so in practice as well as benchmarks. Chrome is significantly faster than Firefox, Opera and IE. I haven't tested Safari though.

I'm getting between 115K and 135K nps on Safari 5 on a 2.66 GHz Core2Duo. Still loses to a Forth TSCP clone getting about the same depth and nps. Seems like there are some bugs in your search. It is not seeing tactics and mates in accordance with the depth reported.

(Is there a way to make your test harness play white? Buttons to switch sides or undo a move would be nice.)

[Albert Silver]

According to latest tests, Chrome 6 is the king of scripts.

It certainly seems so in practice as well as benchmarks. Chrome is significantly faster than Firefox, Opera and IE. I haven't tested Safari though.

I'm getting between 115K and 135K nps on Safari 5 on a 2.66 GHz Core2Duo.

How does it compare to Chrome 6?

[gladius]

I'm getting between 115K and 135K nps on Safari 5 on a 2.66 GHz Core2Duo. Still loses to a Forth TSCP clone getting about the same depth and nps. Seems like there are some bugs in your search. It is not seeing tactics and mates in accordance with the depth reported.

(Is there a way to make your test harness play white? Buttons to switch sides or undo a move would be nice.)

I have no doubt there are bugs in the search . The other issue is it's using some pretty heavy pruning, and a dumb q-search, so it's depths aren't really comparable to a TSCP type program. I just discovered the ply reported was 1 higher than it actually was reaching as well. Also, TSCP has a fair bit more eval. knowledge if I remember correctly, which can make a big difference.

I'm working on fixing some bugs (eval/search) currently. Adding undo and the ability to play either color is definitely on the list for the next release as well.

Thanks for the comments!

[gladius]

One suggestion: Display the NPS as an integer, without the fractional part.

Excellent suggestion, done . It'll be out in the next release.