Not logged inRybka Chess Community Forum
Up Topic Rybka Support & Discussion / Rybka Discussion / Don't understand this behaviour of Rybka 3
- - By Amit Bhave (*) Date 2010-09-30 16:40
FEN:
8/1p1n4/4kp2/PK2p1p1/1B2P1P1/7P/8/8 w - - 1 0


It is the position from the game Anand,Viswanathan (2800) - Hammer,Jon Ludvig (2636) at Arctic Stars Prelim Kristiansund 2010.

Rybka 3 anaylsis gives:

Rybka 3:
   2  00:00           114  116.736  +0.40  1.Be1
   3  00:00           151  154.624  +0.36  1.Be1
   4  00:00           253  259.072  +0.39  1.Be1
   5  00:00           421  26.944  +0.39  1.Be1 Ke7
   6  00:00           506  32.384  +0.37  1.Be1 Ke7 2.Bf2
   7  00:00           693  44.352  +0.36  1.Be1 Ke7 2.Bf2 Ke6
   8  00:00         1.299  83.136  +0.39  1.Be1 Ke7 2.Bf2 Ke6 3.Be3
   9  00:00         1.744  111.616  +0.39  1.Be1 Ke7 2.Bf2 Ke6 3.Be3 Kd6
  10  00:00         2.529  161.856  +0.38  1.Be1 Ke7 2.Bf2 Ke6 3.Be3 Kd6 4.Bg1
  11  00:00         4.136  132.352  +0.39  1.Be1 Ke7 2.Bf2 Ke6 3.Be3 Kd6 4.Bg1
  12  00:00         5.314  115.777  +0.39  1.Be1 Ke7 2.Bf2 Ke6 3.Be3 Kd6 4.Bg1
  13  00:00         7.629  124.001  +0.39  1.Be1 Ke7 2.Bf2 Ke6 3.Be3 Kd6 4.Bg1 Ke6 5.Bb6 Nb8
  14  00:00        11.953  130.211  +0.39  1.Be1 Ke7 2.Bf2 Ke6 3.Be3 Kd6 4.Bg1 Kc7 5.Be3 Nb8
  15  00:00        15.580  127.631  +0.39  1.Be1 Ke7 2.Bf2 Ke6 3.Be3 Kd6 4.Bg1 Kc7
  16  00:00        20.376  121.308  +0.39  1.Be1 Ke7 2.Bf2 Ke6 3.Be3 Kd6 4.Bg1 Kc7
  17  00:00        27.491  119.790  +0.39  1.Be1 Ke7 2.Bf2 Ke6 3.Be3 Kd6 4.Bg1 Kc7 5.Bf2 Kd6 6.Be3 Kc7 7.Bf2 Kd6 8.Be3 Kc7 9.Bf2 Kd6 10.Be3 Kc7 11.Bf2 Kd6 12.Be3 Kc7 13.Bf2 Kd6 14.Be3 Kc7 15.Bf2 Kd6 16.Be3
  18  00:00        39.625  123.707  +0.39  1.Be1 Ke7 2.Bf2 Ke6 3.Be3 Kd6 4.Bg1 Kc7 5.Bf2 Kd6 6.Be3 Kc7 7.Bf2 Kd6 8.Be3 Kc7 9.Bf2 Kd6 10.Be3 Kc7 11.Bf2 Kd6 12.Be3 Kc7 13.Bf2 Kd6 14.Be3 Kc7 15.Bf2 Kd6 16.Be3
  18+  00:00        77.358  137.049  +0.59  1.a6
  18+  00:01       123.076  136.840  +1.19  1.a6
  18  00:01       141.327  142.580  +0.39  1.Be1 Ke7 2.Bf2 Ke6 3.Be3 Kd6 4.Bg1 Kc7 5.Bf2 Kd6 6.Be3 Kc7 7.Bf2 Kd6 8.Be3 Kc7 9.Bf2 Kd6 10.Be3 Kc7 11.Bf2 Kd6 12.Be3 Kc7 13.Bf2 Kd6 14.Be3 Kc7 15.Bf2 Kd6 16.Be3
  19  00:01       175.713  137.246  +0.39  1.Be1 Ke7 2.Bf2 Ke6 3.Be3 Kd6 4.Bg1 Kc7 5.Bf2 Kd6 6.Be3 Kc7 7.Bf2 Kd6 8.Be3 Kc7 9.Bf2 Kd6 10.Be3 Kc7 11.Bf2 Kd6 12.Be3 Kc7 13.Bf2 Kd6 14.Be3 Kc7 15.Bf2 Kd6 16.Be3
  20  00:02       321.286  151.681  +0.39  1.Be1 Ke7 2.Bf2 Ke6 3.Be3 Kd6 4.Bg1 Kc7 5.Bf2 Kd6 6.Be3
  21  00:03       463.834  157.743  +0.39  1.Be1 Ke7 2.Bf2 Ke6 3.Be3 Kd6 4.Bg1 Kc7 5.Bf2 Kd6 6.Be3
  22  00:04       660.933  161.256  +0.39  1.Be1 Ke7 2.Bf2 Ke6 3.Be3 Kd6 4.Bg1 Kc7 5.Bf2 Kd6 6.Be3
  23  00:06     1.118.525  169.936  +0.39  1.Be1 Ke7 2.Bf2 Ke6 3.Be3 Kd6 4.Bg1 Kc7 5.Bf2 Kd6 6.Be3
  23+  00:07     1.333.330  172.281  +0.59  1.a6
  23+  00:13     2.279.728  168.126  +3.59  1.a6

Question: What is this 18+ move? why did Rybka switch back from the winning move (a6) on this 18+ depth to start with Be1 and then switch back to a6?

Regards
A
Parent - By Felix Kling (Gold) Date 2010-09-30 16:42
rybka sees a6 is winning at depth 18, but then the score drops again for that move, only later she sees that it's really winning.
Parent - - By Jury Osipov (**) Date 2010-10-01 10:21
Alles ist in Ordnung.

There are two reasons:
1. Aspiration search in the root.
2. Rewriting of PV in hash (maybe this is bug).
Parent - - By Amit Bhave (*) Date 2010-10-01 10:27
Hi,
I had cleared the hash several times before running the analysis.
In few attempts, a6 was never analysed by Rybka. In others, you got the analysis as shown.

Is this hash-bug something that has been resolved in Rybka 4 ?

Thanks,
A
Parent - By kunah (**) Date 2010-10-01 11:29
Well I don't think it's a bug. Even if you clear the hashtable, there are other things like the history table; so yes, it's possible that the results differ a bit now and then (especially if you have more than one candidate moves which score similarly). This also explains why when aspiration window fails (read low or high), you don't research exactly the same three with a wider window so there can be some minor fluctuations. Overwriting PV in hash table (considering node count and previous hash table clear) seems unlikely to me. Maybe some effects of parallel search? Hard to guess. Did you try to restart the engine?
Up Topic Rybka Support & Discussion / Rybka Discussion / Don't understand this behaviour of Rybka 3

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill