Bug in Rybka 3 ?

By Eelco de Groot Date 2008-09-25 04:57

[quote="Orlando de la Vega"]
Hi guys

,

please can some of you have a look at this position ?

Now make an analyze with Rybka 3 ! How many times does it think in 12th proof ???
I count 4 times with my Rybka 3 1-CPU !!!

Bug or not ?

Best,
Orlando
[/quote]

Hello Orlando,

Could you maybe explain what you mean with thinking in 12th proof?

Something like 12 ply? What GUI did you use, or maybe post the output?
Have not got Rybka 3 so I can't take a look. But still hoping there will be some
sort of bugfix for Rybka 3 one day..

Eelco

By Orlando de la Vega Date 2008-09-25 12:28

> Could you maybe explain what you mean with thinking in 12th proof?

> Something like 12 ply? What GUI did you use, or maybe post the output?

Hi Elco,

what I meant is when I do an analyze with Rybka 3 on this position, I see that it does 4 loops on depth 12 !

Lets have a look at its PVs in console mode.

Code:

position fen 1r5k/p1q1p2p/3p1r2/P2B1P1Q/2pB4/1p6/2P3PP/R6K b - -
go infinite
info depth 1
info depth 2
info depth 2 score cp -72 time 1 nodes 148 nps 151552 pv b3c2
info depth 2 time 16 nodes 152 nps 9728
info depth 3
info depth 3 score cp -70 time 16 nodes 258 nps 16512 pv b3c2
info depth 3 time 32 nodes 264 nps 8448
info depth 4
info depth 4 time 48 nodes 620 nps 13226
info depth 4 score cp -68 time 79 nodes 949 nps 12300 pv b3c2
info depth 4 time 79 nodes 949 nps 12300
info depth 5
info depth 5 time 110 nodes 1438 nps 13386
info depth 5 score cp -84 time 126 nodes 1643 nps 13352 pv b3c2 a1c1
info depth 5 time 141 nodes 1644 nps 11939
info depth 6
info depth 6 score cp -93 time 172 nodes 2372 nps 14121 pv b3c2 a1c1 c7a5
info depth 6 time 188 nodes 2522 nps 13736
info depth 7
info depth 7 score cp -96 time 235 nodes 3662 nps 15956 pv b3c2 a1c1 c7a5 h5f7
info depth 7 time 266 nodes 3988 nps 15352
info depth 8
info depth 8 score cp -85 time 578 nodes 12752 nps 22591 pv b3c2 d5e4 c7a5 e4c2
a5d2 h5d1
info depth 8 time 609 nodes 13063 nps 21964
info depth 9
info depth 9 time 1171 nodes 26575 nps 23238
info depth 9 score cp -99 time 1389 nodes 33166 nps 24450 pv b3c2 d5e4 c7a5 e4c2
 a5d2 h5d1 d2d1 c2d1 a7a5 d1e2 b8b4
info depth 9 time 1405 nodes 33167 nps 24172
info depth 10
info depth 10 time 2419 nodes 52475 nps 22213
info depth 10 score cp -130 time 3183 nodes 69763 nps 22443 pv b3c2 a1c1 c7a5 h5
f7 a5d8 d5e4
info depth 10 time 3246 nodes 70705 nps 22304
info depth 11
info depth 11 score cp -130 time 5508 nodes 136521 nps 25380 pv b3c2 a1f1 c7a5 h
5f7 a5d8 d5e4 b8b1 f1b1 c2b1q
info currmove c7a5 currmovenumber 2
info currmove b3b2 currmovenumber 3
info currmove c4c3 currmovenumber 4
info currmove a7a6 currmovenumber 5
info currmove e7e5 currmovenumber 6
info currmove e7e6 currmovenumber 7
info currmove h7h6 currmovenumber 8
info currmove c7c5 currmovenumber 9
info currmove c7b6 currmovenumber 10
info currmove c7c6 currmovenumber 11
info currmove c7b7 currmovenumber 12
info currmove c7d7 currmovenumber 13
info currmove c7c8 currmovenumber 14
info currmove c7d8 currmovenumber 15
info currmove b8b4 currmovenumber 16
info currmove b8b5 currmovenumber 17
info currmove b8b6 currmovenumber 18
info currmove b8b7 currmovenumber 19
info currmove b8a8 currmovenumber 20
info currmove b8c8 currmovenumber 21
info currmove b8d8 currmovenumber 22
info currmove b8e8 currmovenumber 23
info currmove b8f8 currmovenumber 24
info currmove b8g8 currmovenumber 25
info currmove h8g7 currmovenumber 26
info depth 11 time 5976 nodes 139277 nps 23865
info depth 12
info currmove b3c2 currmovenumber 1
info currmove c7a5 currmovenumber 2
info currmove b3b2 currmovenumber 3
info currmove c4c3 currmovenumber 4
info currmove a7a6 currmovenumber 5
info currmove e7e5 currmovenumber 6
info currmove e7e6 currmovenumber 7
info currmove h7h6 currmovenumber 8
info currmove c7c5 currmovenumber 9
info currmove c7b6 currmovenumber 10
info currmove c7c6 currmovenumber 11
info currmove c7b7 currmovenumber 12
info currmove c7d7 currmovenumber 13
info currmove c7c8 currmovenumber 14
info currmove c7d8 currmovenumber 15
info currmove b8b4 currmovenumber 16
info currmove b8b5 currmovenumber 17
info currmove b8b6 currmovenumber 18
info currmove b8b7 currmovenumber 19
info currmove b8a8 currmovenumber 20
info currmove b8c8 currmovenumber 21
info currmove b8d8 currmovenumber 22
info currmove b8e8 currmovenumber 23
info currmove b8f8 currmovenumber 24
info currmove b8g8 currmovenumber 25
info currmove h8g7 currmovenumber 26
info depth 12 time 12434 nodes 292713 nps 24106
info currmove b3c2 currmovenumber 1
info currmove c7a5 currmovenumber 2
info currmove b3b2 currmovenumber 3
info currmove c4c3 currmovenumber 4
info currmove a7a6 currmovenumber 5
info currmove e7e5 currmovenumber 6
info currmove e7e6 currmovenumber 7
info currmove h7h6 currmovenumber 8
info currmove c7c5 currmovenumber 9
info currmove c7b6 currmovenumber 10
info currmove c7c6 currmovenumber 11
info currmove c7b7 currmovenumber 12
info currmove c7d7 currmovenumber 13
info currmove c7c8 currmovenumber 14
info currmove c7d8 currmovenumber 15
info currmove b8b4 currmovenumber 16
info currmove b8b5 currmovenumber 17
info currmove b8b6 currmovenumber 18
info currmove b8b7 currmovenumber 19
info currmove b8a8 currmovenumber 20
info currmove b8c8 currmovenumber 21
info currmove b8d8 currmovenumber 22
info currmove b8e8 currmovenumber 23
info currmove b8f8 currmovenumber 24
info currmove b8g8 currmovenumber 25
info currmove h8g7 currmovenumber 26
info depth 12 time 16521 nodes 376317 nps 23324
info currmove b3c2 currmovenumber 1
info currmove c7a5 currmovenumber 2
info currmove b3b2 currmovenumber 3
info currmove c4c3 currmovenumber 4
info currmove a7a6 currmovenumber 5
info currmove e7e5 currmovenumber 6
info currmove e7e6 currmovenumber 7
info currmove h7h6 currmovenumber 8
info currmove c7c5 currmovenumber 9
info currmove c7b6 currmovenumber 10
info currmove c7c6 currmovenumber 11
info currmove c7b7 currmovenumber 12
info currmove c7d7 currmovenumber 13
info currmove c7c8 currmovenumber 14
info currmove c7d8 currmovenumber 15
info currmove b8b4 currmovenumber 16
info currmove b8b5 currmovenumber 17
info currmove b8b6 currmovenumber 18
info currmove b8b7 currmovenumber 19
info currmove b8a8 currmovenumber 20
info currmove b8c8 currmovenumber 21
info currmove b8d8 currmovenumber 22
info currmove b8e8 currmovenumber 23
info currmove b8f8 currmovenumber 24
info currmove b8g8 currmovenumber 25
info currmove h8g7 currmovenumber 26
info depth 12 time 26724 nodes 582264 nps 22310
info currmove b3c2 currmovenumber 1
info depth 12 score cp -218 time 40265 nodes 917092 nps 23323 pv b3c2 a1g1 c7a5
h5f7
info currmove c7a5 currmovenumber 2
info currmove b3b2 currmovenumber 3
info currmove c4c3 currmovenumber 4
info currmove a7a6 currmovenumber 5
info currmove e7e5 currmovenumber 6
info currmove e7e6 currmovenumber 7
info currmove h7h6 currmovenumber 8
info currmove c7c5 currmovenumber 9
info currmove c7b6 currmovenumber 10
info currmove c7c6 currmovenumber 11
info currmove c7b7 currmovenumber 12
info currmove c7d7 currmovenumber 13
info currmove c7c8 currmovenumber 14
info currmove c7d8 currmovenumber 15
info currmove b8b4 currmovenumber 16
info currmove b8b5 currmovenumber 17
info currmove b8b6 currmovenumber 18
info currmove b8b7 currmovenumber 19
info currmove b8a8 currmovenumber 20
info currmove b8c8 currmovenumber 21
info currmove b8d8 currmovenumber 22
info currmove b8e8 currmovenumber 23
info currmove b8f8 currmovenumber 24
info currmove b8g8 currmovenumber 25
info currmove h8g7 currmovenumber 26
info depth 12 time 41638 nodes 934124 nps 22972
info depth 13
info currmove b3c2 currmovenumber 1

So I see here that Rybka 3 calculate 4 times on depth 12 !!! Do you understand ?

Best,
Orlando

By Eelco de Groot Date 2008-09-26 08:20 Edited 2008-09-26 08:24

Thanks Orlando, that makes it much more clear!

It could really be a bug, but also Rybka's search seems to just continue normally and the nodecount is going up. Unfortunately the nodecount in Rybka does not tell us very much, not even if the search is actually doing something

But assuming it is not a bug then this kind of behaviour you would expect to pop up in many positions, not just in this one that you found Orlando. In Ancalagon, (a modified Glaurung) there is actually something similar because every search is started before plydepth = 1 with plydepths that are smaller than 1, so with negative depths -3, -2, -1, 0. Tord's Glaurung automatically translates negative plydepths to searchdepth = 0 which is equal to calling the quiescence search, the search that only makes capturing moves or checks in a position so the quiescence search tries to find a quiet position, where eval works better. Calling the qsearch several times does not have any real purpose but later on when the searchnumbers, the plydepths, go up and then calling the search first with smaller plydepths can have some purpose by achieving better move ordering. There were some hints in the Rybka forum that Rybka does repeat searches in some ways so that is basically why I started experimenting with this. But I have no real idea how all this is implemented in Rybka, and I have not heard anybody else who reported this existing in Rybka 1.0 Beta or in the Strelka clones. Basically all we know is that Rybka also starts with negative plydepths and I suspect it is an important part of Rybkas search.

It does not really explain why only here in this position you would see a repetition in Rybkas output, that is strange, but possibly this is not a bug!

Regards, Eelco

By Benno Hartwig Date 2008-09-26 08:48

Code:

So I see here that Rybka 3 calculate 4 times on depth 12 !!! Do you understand ?

Wow, that looks interesting.

As far as I know most time Vas is reading the postings in
http://rybkaforum.net/cgi-bin/rybkaforum/board_show.pl?bid=2

It could be a good idea to post there what you saw.
I assume, Vas will give you an answer, an explanation, or the info, that this is not the behaviour he wanted.

Benno

By Daniel Mehrmann (CSS-Forum) Date 2008-09-26 09:00 Edited 2008-09-26 09:03

Hello Benno,

LOL, you're funny.

If it's no bug Vas will make no post and explain nothing as he does it every time.
If it's a bug, i guess he doesn't write much at all und Conveka might release a bugfix one day.

Don't understand me wrong. He must hide his secrets, that's how buisness works.

Best,
Daniel

By Benno Hartwig Date 2008-09-26 09:44

Code:

LOL, you're funny.

fine, that you are in good spirits.

Code:

If it's no bug Vas will make no post and explain nothing as he does it every time.

No, Vas often has posted there. And especial if he thinks that there is no bug, he will explain it.

Code:

If it's a bug, i guess he doesn't write much at all und Conveka might release a bugfix one day.

OK, sometimes he is quiet, if there is spoken of concrete 'bugs' und ugly 'features' again and again. (wrong bishop, bishop-underpromotion...)
Has there been a new bug, Vas hasn't have answered to?
I shurely assume (funny?), Vas will answer even in this case.

If you want an answer from Vas, ask Vas, do it in Vas' Forum.

Benno

By Jens Heerklotz Date 2008-09-26 21:53

Its not a bug. Let Sjeng 3.0 run on a quadcore and in many positions you watch him do a similar thing (not in exactly the same manner but very much alike). It has to do with smp search implementation. Oh btw sjeng 3,0 was out before rybka 3, just to say it before somebody says "sjeng 3,0 does it also, sjeng is a clone of rybka"

)) Nonono sjeng 3 was out a lot earlier and had it first... so if an engine is a clone then rybka cloned this from sjeng