Hydra Can be Easily Beaten, Sometimes.

[Tony Thomas]

Code: Select all

[Event "?"]
[Site "chessclub.com"]
[Date "2008.5.29"]
[Time "5:50:22"]
[Round "?"]
[White "chess-rehab"]
[Black "apocalypse1112"]
[Result "1-0"]
[WhiteElo "2259"]
[BlackElo "2377"]
[Opening "A00 Anti-Borg (Desprez) opening"]

{Time control = 3 0 Engine book(s):  }
1. h4 e5 {0.00/1 0.0 } 2. c3 Nc6 {0.00/1 5.1 } 3. g3 Nf6 {1.28/12 3.5 } 4. d4 d5
{0.00/1 4.8 } 5. Bg5 exd4 {0.00/1 4.3 } 6. cxd4 h6 {0.96/12 3.8 } 7. Bf4 Bb4+
{0.00/1 4.3 } 8. Nc3 Nh5 {0.84/11 3.9 } 9. e3 Nxf4 {0.91/11 3.9 } 10. gxf4 O-O
{0.00/1 4.0 } 11. Bd3 Re8 {1.31/11 3.2 } 12. Nf3 Bg4 {1.34/11 3.8 } 13. Be2 Qf6
{1.39/10 5.6 } 14. Ne5 Bxe2 {0.00/1 3.5 } 15. Qxe2 Bxc3+ {1.18/12 5.8 } 16. bxc3
Nxe5 {0.00/1 4.0 } 17. fxe5 Qe7 {0.00/1 5.8 } 18. Qg4 Qa3 {1.13/12 3.5 } 19. O-O
c5 {0.00/1 3.5 } 20. Kh2 Qxc3 {0.00/1 3.1 } 21. Rg1 g6 {0.95/11 3.2 } 22. h5 g5
{0.00/1 3.9 } 23. Qf5 Rac8 {0.94/10 3.0 } 24. Rxg5+ Kf8 {-2.05/12 2.4 } 25. Rg7
Rc7 {0.00/1 3.4 } 26. Rag1 Qb2 {-M 04/7 0.2 } 27. Rg8+ {apocalypse1112 forfeits
on time} 1-0

Romi resigned after move 27 but I didnt enter resign so she lost on time...Anything unusual about this game??

I am thinking that it is an LMR combined with null move problem. Romi 3k reduces 2 ply for LMR. Then a null move reduction in an already reduced line may just simply overlook the mate threat and prune the line as good for the computer. I will try not allowing null moves in LMR reduced lines and see what happens.

Holy goat Tony! It seems like it's working. Romi is more sure footed than Goat herself. Goat has even asked Romi to give stepping lessons to all the kidds of goat mountain!

This was just a quickie. I will see if there is more to be gotten from better design and testing.

Are you saying that Romi now sees the problem much faster? I was so surprised to see Romi getting mauled by a human in such a fast time control. Also, some work to increase compatibility with the chesspartner would be appreciated. I know that its simply a waste of time, but losing on time only ensures us the last spot in ACCA which is reserved for Swami and his Twisted Logic running on his new Q6600.. By the way, you never answered my question in the email. Approximately how much faster is Romi on a 64bit platform?? How about a quick and dirty copy/paste type implementation of SMP from Viper?? Since your program is open source, you can add the disclaimer below to the read me. I might be on to something regards..
Jealous poor American regards on top of the early regards, regards
Tony

This program is an offspring of the copulation of ideas I got from reading the source code of other chess engines out there.

[swami]

Hiarcs has joined...soon will be followed by Rybka, Naum, Zappa, Thinker, Learning-On Romi etc

now here goes my chances of winning

[Michael Sherwin]

A non pgo compile on a Q6600 at 2.4 GHz, opening position:

32 bit Romi, ~3,000,000 nps

64 bit Romi, ~3,600,000 nps

That's about 10 CST2 faster!

more accurate measurements to follow as well as pgo compile rates.

[Tony Thomas]

I dont have a calculator, looks to me like about 18% faster, is that right? So, you have a Q6600 now?? Why dont you just run Romi then?

[Michael Sherwin]

I dont have a calculator, looks to me like about 18% faster, is that right? So, you have a Q6600 now?? Why dont you just run Romi then?

A friend has one ...

Memory, my bad!

Here are the correct numbers.

Code: Select all

opening position:
32 bit     2,470,407
32 bit pgo 2,544,241
64 bit     3,075,535
64 bit pgo 3,322,854

after e4 e5 Nf3 Nc6 Bc4 Bc5 0-0 Nf6 d3 0-0 Nc3 d6
32 bit     2,779,078
32 bit pgo 2,932,710
64 bit     3,459,875
64 bit pgo 3,837,569

... and he's not very interested in chess.

[Tony Thomas]

Wow, my reply just mysteriously disappeared. Where is the darn stimulus when I need it badly. I wish I had friends in my neighbourhood with Q6600.