Deep Junior 13 Welcome Test

Discussion of anything and everything relating to chess playing software and machines.

Moderators: hgm, Rebel, chrisw

beram
Posts: 1187
Joined: Wed Jan 06, 2010 3:11 pm

Re: Deep Junior 13 Welcome Test

Post by beram »

That preferred match of Stfish- Houdini would be a very interesting tactical match I think

As far for the games DJ played I found one strange loss in the first game of DJ against Stockfish.
After 37. Qf4 ? DJ is suddenly lost but after 37. g6 or Qg7 the result would be a draw because after that Rxc1 is forced and black must take the draw.
Does anyone has an explanation ?
[D] 4knR1/5p2/p2p1Q2/1p1q2P1/4p3/P7/1Pr5/1KB5 w - - 0 37
Jouni
Posts: 3293
Joined: Wed Mar 08, 2006 8:15 pm

Re: Deep Junior 13 Welcome Test

Post by Jouni »

Yes very strange, because there are 4 drawing moves according to Critter 1.4 64-bit:

1. = (0.00): 37.Qh6 Rxc1+ 38.Kxc1 Qc4+ 39.Kb1 Qd3+ 40.Ka2 Qd5+ 41.Kb1 Qd1+ 42.Ka2 Qd5+
2. = (0.00): 37.Qh8 Rxc1+ 38.Kxc1 Qc4+ 39.Kb1 Qd3+ 40.Ka2 Qd5+ 41.Kb1 Qd1+ 42.Ka2 Qd5+
3. = (0.00): 37.Qg7 Rxc1+ 38.Kxc1 Qc4+ 39.Kb1 Qd3+ 40.Ka1 Qd1+ 41.Ka2 Qd5+ 42.Kb1 Qd1+ 43.Ka2
4. = (0.00): 37.g6 Rxc1+ 38.Kxc1 Qc4+ 39.Kd2 Qd3+ 40.Ke1 Qe3+ 41.Kf1 Qd3+ 42.Kg2 Qd2+ 43.Kh3 Qe3+ 44.Kh4 Qe1+ 45.Kh5 Qh1+ 46.Kg5 Qc1+ 47.Qf4 f6+ 48.Kg4 Qd1+ 49.Kh4 Qh1+ 50.Kg3 Qg1+ 51.Kh3 Qh1+ 52.Qh2 Qd1 53.Qg3 Qh5+ 54.Kg2 Qe2+ 55.Qf2 Qg4+ 56.Qg3 Qe2+

Junior evaluates +0,35 after 24 ply and 5:50! Also against Houdini there was at least one directly losing move - someone speculates about hash collisions.

Jouni
Jouni
Posts: 3293
Joined: Wed Mar 08, 2006 8:15 pm

Re: Deep Junior 13 Welcome Test

Post by Jouni »

So why is test stopped? Without stop we would have already result for 60 games :( If You change conditions, please note, that Critter fills 1 GB has in 20 sec, so if there is more RAM allocate it!

Jouni
TimoK
Posts: 98
Joined: Sun Jan 03, 2010 12:28 pm
Location: Hamburg

Re: Deep Junior 13 Welcome Test

Post by TimoK »

Hi Jouni,

the test will continue very soon (probably today or tomorrow evening). You are right about the hash, some Junior-experts told me, that it also uses more hash when playing with >2 threads. So the new test will give 4GB hash to all engines. Time control will also be a bit adjusted.
I'll post the hyperlinks to the new test as soon as the test will have been started.

Stay tuned and best regards,
Timo

P.S.: I reported the game / position where Junior played these horrible blunder to Amir Ban (along with 2 other games where something similar happened). Maybe he can find the bug.