Optimizing Hiarcs 12 MP

PortCitySlim · Post by **PortCitySlim** » Tue Apr 01, 2008 5:31 am

In the playchess engine room I have heard that Rybka is winning over 60% of the games being played on equal equipment. I wont have my copy of Deep H12 till later this week, however, I would like to know if you guys have any advice on how to configure the program to at least give Rybka a fight at 40/20 time controls down to 4 or 5 min games. I am running Dual 2.2ghz and 4gb of ram on Vista Ultimate(64bit). thanks

Dylan Sharp · Post by **Dylan Sharp** » Tue Apr 01, 2008 6:34 am

(Reply reposted below as the posts appeared at the same time and maybe he overlooked it.)

PortCitySlim · Post by **PortCitySlim** » Tue Apr 01, 2008 6:34 am

Since H12 was released I have spent a few hours watching games played on the playchess server at various time controls to try and help me make up my mind if I should buy yet another HIARCS 32bit engine ( Latest one I have is 11.2 ) ... I know it was not a very scientific approach but I only saw H12 win 1 game!! some draws and plenty of losses ... Of course books etc can make a big difference but by watching plenty of games you start to get a feel and it feels like H12 will not even make it to number 2 on the rankings

This is one example of the above statements I was talking about, this quote came from the Rybka forum.

Dylan Sharp · Post by **Dylan Sharp** » Tue Apr 01, 2008 6:37 am

My advice is to work on the book, as what some people in there do when they have a weaker hardware is to build a very strong book, and then book win the opponent, or at least have a winning advantage on time after getting out of book.

Focus in some obscure but strong variation that it's not on the book of the opponents, and build your strategy out of it, by deeply analyzing most of its variations so you're ready against everything, and these openings at fast time controls are played instantly backed up by your offline analysis at slow time controls.

Good luck.

PortCitySlim · Post by **PortCitySlim** » Tue Apr 01, 2008 6:41 am

Thanks Dylan, are there any setting that help H12 play better at diffrent time controls? On setting I have heard a little about is the Sharpen PV setting?

Dylan Sharp · Post by **Dylan Sharp** » Tue Apr 01, 2008 11:31 am

Sharpen PV ON for longer time controls and OFF for quick time controls is the only I'm aware of. I suppose the default settings are the strongest.

billyraybar · Post by **billyraybar** » Tue Apr 01, 2008 2:17 pm

Are there any indications that turning futility pruning off helps at long time controls? I've been toying around with it but I haven't been able to see if it helps or hurts yet

PortCitySlim · Post by **PortCitySlim** » Tue Apr 15, 2008 7:15 am

After 51 fast(3m0s,3m1s,5m0s) games in the playchess engine room my Deep HIARCS 12 elo is 2470. I am running on a AMD X2 Turion TL 64 2.2ghz(2048mb hash) which is slightly slower then the average machine in the engine room. 42 of my games were against Rybka engines and I have a 20.5/51(40.2%) result with an average opponents elo of 2552. I used default settings with book set to optimize. I did change the Forward Pruning from FP10 to FP11 for ten games, which I felt improved the play for speed games(still unsure though). I have been considering changing the Threat Depth(to maybe 4 or 6) and possibly taking off hyper-modern play for speed games in hopes of improving HIARCS strength in 5min and less games. Does anyone have any tips about these setting for speed games?

I feel that the only major weakness of HIARCS is its endgame strength, after watching 51 games I saw HIARCS lose several endgames where it was evaluating positively and most of the time up material and never saw HIARCS comeback when negative or down material. HIARCS had a slight advantage after the opening in most games probably due to its super strong book and I feel that is HAIRCS's main advantage.

Post by **Harvey Williamson** » Tue Apr 15, 2008 2:35 pm

The default settings should be the best for blitz but if you find otherwise please let us know.

Playchess although it is great fun is not the most reliable place to test as each game is v different hardware, different book, different engine etc...

PortCitySlim · Post by **PortCitySlim** » Wed Apr 16, 2008 7:41 am

Ok, after 15 more games with FP11 setting H12 improved its elo to 2551(engine room) and the endgame problems have disappeared. I retract what I said before about H12's endgame ability, I just didn't give it enough time before. H12's book is amazing and I haven't seen it lose with white against the Poison Pawn Najdorf. I still haven't played enough games to be sure but it looks like FP11 is helping my H12 at blitz time controls. Here is a game H12 played against a Rybka that had me out-powered(I think it was Intel Dual 2.66-3.0ghz) where the fishka was winning(as much as 3.23) until the fatal 46..a5?? and H12 capitalized to win even down 1 pawn and facing 2 passed queenside pawns.

[Event "Rated game, 3m + 0s"]
[Site "Engine Room"]
[Date "2008.04.15"]
[White "PortCitySlim, Deep Hiarcs 12"]
[Black "Yuksel09, Rybka 2.3.2a mp 32-"]
[Result "1-0"]
[ECO "D43"]
[WhiteElo "2531"]
[BlackElo "2601"]

{DHiarcs12.ctg, 2048 MB} 1. d4 {B/0 0} Nf6 {0} 2. Nf3 {B/0 0} d5 {(g6) 0} 3. c4
{B/0 0} e6 {(c6) 0} 4. Nc3 {B/0 0} c6 {0} 5. Bg5 {B/0 0} h6 {(dxc4) 0} 6. Bh4 {
B/0 0} dxc4 {0} 7. e4 {B/0 0} g5 {0} 8. Bg3 {B/0 0} b5 {0} 9. Be2 {B/0 0} Bb7 {
(Bb4) 0} 10. h4 {B/0 0} g4 {(b4) 0} 11. Ne5 {B/0 0} Nbd7 {0} 12. Bxg4 {B/0 0}
Rg8 {(Nxe5) 0} 13. Nxd7 {B/0 0} Qxd7 {0} 14. Bf3 {B/0 0} O-O-O {0} 15. Qd2 {
B/0 0} c5 {0} 16. Qf4 {B/0 0} Rxg3 {0} 17. fxg3 {B/0 0} cxd4 {0} 18. O-O-O {
B/0 0} Ne8 {0} 19. e5 {B/0 0} Bxf3 {(Ba6) 0} 20. gxf3 {0.12/13 13} Nc7 {0} 21.
a3 {0.04/12 12} d3 {0} 22. g4 {-0.20/12 6} Nd5 {-0.33/12 4} 23. Nxd5 {
-0.49/12 5} Qxd5 {-0.38/13 3} 24. Rh2 {-1.17/12 11} Bc5 {(Rd7) -0.86/15 3} 25.
Rhd2 {-1.34/12 7} Rd7 {-0.96/10 0} 26. Kb1 {-1.55/13 7} Bxa3 {-1.11/14 3} 27.
bxa3 {-1.55/13 1} c3 {-1.75/15 10} 28. Rb2 {-1.86/12 0} c2+ {-2.49/13 4} 29.
Kc1 {-1.91/13 1} cxd1=Q+ {-2.56/14 1} 30. Kxd1 {-2.20/14 0} Rc7 {-2.65/15 2}
31. Qe3 {-3.20/13 6} Rc3 {(h5) -2.67/15 0} 32. Rd2 {-3.23/12 8} Rxa3 {
(h5) -2.79/15 8} 33. Ke1 {-2.52/11 1} Kb7 {-2.83/14 5} 34. Kf2 {-2.77/13 1} Ra2
{-3.36/12 3} 35. Ke1 {-2.30/13 1} Ra1+ {(Rxd2) -2.09/13 18} 36. Kf2 {-2.12/13 3
} Ra3 {-1.95/14 1} 37. Qf4 {-2.78/13 8} Ka6 {-2.67/10 3} 38. Qb4 {-2.84/13 9}
Ra2 {-2.74/12 1} 39. Rxa2+ {-1.87/13 3} Qxa2+ {-2.74/14 1} 40. Ke3 {-1.87/14 0}
Qe2+ {-1.69/14 12} 41. Kd4 {-1.80/15 0} Qf2+ {(Qc2) -1.69/14 4} 42. Ke4 {
-1.41/13 2} Qc2 {-1.06/15 8} 43. Ke3 {-1.28/14 0} Qc1+ {-1.00/14 5} 44. Kxd3 {
-0.82/13 0} Qd1+ {(Qc6) -1.19/12 2} 45. Ke3 {-0.88/12 1} Kb6 {-1.04/12 1} 46.
h5 {-0.59/13 6} a5 {(Qg1+) -0.08/13 3} 47. Qd4+ {7.87/14 5} Qxd4+ {10.18/12 8}
48. Kxd4 {9.87/18 0} Kc6 {(a4) 10.19/12 1} 49. g5 {11.14/13 4} hxg5 {
(a4) 10.43/13 0} 50. h6 {
Yuksel09,Rybka 2.3.2a mp 32-b resigns (Lag: Av=0.38s, max=0.9s) 11.16/11 1}
1-0

PortCitySlim · Post by **PortCitySlim** » Sat May 03, 2008 6:46 am

What happens if you reduce the Hiarcs book "variety of play" value under book options to less then "optimize"? I notice that when you select "optimize" book the value goes down greatly but not all the way. This might be a "duh" question but I am inclined to believe that it would make Hiarcs only play the very strongest lines for example if Hiarcs thinks that 1..c5 is the best reply to 1.e4 then it would only play 1..c5 against 1.e4 under those settings? Why would reducing that value a little more hurt?

PortCitySlim · Post by **PortCitySlim** » Tue May 06, 2008 9:12 am

PortCitySlim wrote:What happens if you reduce the Hiarcs book "variety of play" value under book options to less then "optimize"? I notice that when you select "optimize" book the value goes down greatly but not all the way. This might be a "duh" question but I am inclined to believe that it would make Hiarcs only play the very strongest lines for example if Hiarcs thinks that 1..c5 is the best reply to 1.e4 then it would only play 1..c5 against 1.e4 under those settings? Why would reducing that value a little more hurt?

bump. Anyone know the answer to this?

Post by **Harvey Williamson** » Tue May 06, 2008 9:21 am

PortCitySlim wrote:
PortCitySlim wrote:What happens if you reduce the Hiarcs book "variety of play" value under book options to less then "optimize"? I notice that when you select "optimize" book the value goes down greatly but not all the way. This might be a "duh" question but I am inclined to believe that it would make Hiarcs only play the very strongest lines for example if Hiarcs thinks that 1..c5 is the best reply to 1.e4 then it would only play 1..c5 against 1.e4 under those settings? Why would reducing that value a little more hurt?

bump. Anyone know the answer to this?

If you really want to play only c5 then you are best to edit the book directly. Another danger is that it may keep repeating a losing line with no variety.

Dark Horse · Post by **Dark Horse** » Wed Jun 11, 2008 10:39 pm

my Deep HIARCS 12 came today, I'm getting about 1250 Kn/Sec in complex positions and about 1400 - 1600 Kn/Sec in open middle game positions and about 2400 Kn/Sec in end games with a fair amount of material, so I think my hardware is pretty good, I ran a couple of test positions and I am impressed (I use the engine only for analysis and preparation not for engine vs engine matches)

when I run Zappa I get about 1100 Kn/Sec for a position where I get 1400 Kn/Sec with Deep HIARCS, but I get almost 3000 Kn/Sec for the same position with Toga 1.4SE 4 CPU version

so maybe the Kn/Sec is not comparable accross engines ? (I read somewhere that Rybka's Kn/Sec value was not comparable)

but I like HIARCS performance in the test positions, it seems to be able to 0 in on the main line pretty fast while Zappa took a bit of time to do the same thing

Post by **Harvey Williamson** » Wed Jun 11, 2008 10:44 pm

Dark Horse wrote:my Deep HIARCS 12 came today, I'm getting about 1250 Kn/Sec in complex positions and about 1400 - 1600 Kn/Sec in open middle game positions and about 2400 Kn/Sec in end games with a fair amount of material, so I think my hardware is pretty good, I ran a couple of test positions and I am impressed (I use the engine only for analysis and preparation not for engine vs engine matches)

when I run Zappa I get about 1100 Kn/Sec for a position where I get 1400 Kn/Sec with Deep HIARCS, but I get almost 3000 Kn/Sec for the same position with Toga 1.4SE 4 CPU version

so maybe the Kn/Sec is not comparable accross engines ? (I read somewhere that Rybka's Kn/Sec value was not comparable)

but I like HIARCS performance in the test positions, it seems to be able to 0 in on the main line pretty fast while Zappa took a bit of time to do the same thing

Your right you cant compare KN/s between different engines they all count them differently and search to different depths. It is widely acknowledged that both Rybka's depth and KN/s that are displayed are not real. You can probably add 2 or3 to depth and multiply KN/s by 10. I am glad you are enjoying Hiarcs and helping us to defeat the Rybka Wimps.