Topic Rybka Support & Discussion / Rybka Discussion / Rybka plays an antipositional move, Vas can you explain?
![[us]](/mwf/v12/flags/us.png "United States")
Hi,
It's my general understanding that Rybka's positional understanding is very good and is a very useful engine to use for finding good positional moves. I used it to analyze a game I played vs TheKing and saw something curious: after the moves
1.e4 c5
2.Nf3 d6
3.c3 Nf6
4.Be2 Nc6
5.d3 e5
6.0-0 d5
7.exd5 Nxd5
Rybka 2.2 recommends 8.c4 with eval 0.00, depth 17.
That is an obvious antipositional move and I was shocked to see Rybka recommend it. Every other engine I have drops the eval considerably after c4. Is anyone here as shocked as I am?
-Max
It's my general understanding that Rybka's positional understanding is very good and is a very useful engine to use for finding good positional moves. I used it to analyze a game I played vs TheKing and saw something curious: after the moves
1.e4 c5
2.Nf3 d6
3.c3 Nf6
4.Be2 Nc6
5.d3 e5
6.0-0 d5
7.exd5 Nxd5
Rybka 2.2 recommends 8.c4 with eval 0.00, depth 17.
That is an obvious antipositional move and I was shocked to see Rybka recommend it. Every other engine I have drops the eval considerably after c4. Is anyone here as shocked as I am?
-Max
On my system, Rybka 2.2 MP changes to 8.d4 at depth 15 and sticks with it, finding that 8.c4 is -0.01 at that depth.
While Rybka 2.2n2 MP changes to 8.d4 at depth 12 already. If I restrict to one core, it will change to 8.d4 at depth 14.
If we forget about 8.d4, I'm not completely shocked that Rybka does not totally dislike 8.c4. White is playing some Philidor Defense/Benoni hybrid with reversed colors, and Rybka generally is not so pessimistic about these structures as other engines, despite the backwards d-pawn and gaping hole on d4. I do think white is worse after 8.c4, but it's not so easy to explain why this structure is bad, while black is not so much worse in fairly similar Sveshnikov or Kalashnikov structures with a pawn on e4 (or e5) instead of c4 (c5).
After 8.c4 Nf6, my Rybka 2.2n2 MP is now giving -0.06 at depth 19.
While Rybka 2.2n2 MP changes to 8.d4 at depth 12 already. If I restrict to one core, it will change to 8.d4 at depth 14.
If we forget about 8.d4, I'm not completely shocked that Rybka does not totally dislike 8.c4. White is playing some Philidor Defense/Benoni hybrid with reversed colors, and Rybka generally is not so pessimistic about these structures as other engines, despite the backwards d-pawn and gaping hole on d4. I do think white is worse after 8.c4, but it's not so easy to explain why this structure is bad, while black is not so much worse in fairly similar Sveshnikov or Kalashnikov structures with a pawn on e4 (or e5) instead of c4 (c5).
After 8.c4 Nf6, my Rybka 2.2n2 MP is now giving -0.06 at depth 19.
![[hu]](/mwf/v12/flags/hu.png "Hungary")
Yes, I agree with this.
Of course, if you have seen the earlier moves, 8. c4 looks pretty illogical. (And yes, Rybka does eventually reject it after a bigger analysis - see below).
However, the move is also not that horrible objectively, probably second-best by a small margin. Rybka is a bit less dogmatic than other engines about "bad pawn structure" and this is intentional, I like it that way and think that it's appropriate.
Vas
New game - Rybka 2.2 32-bit, 120'/40+120'/40+120'
Analysis by Rybka 2.2 32-bit:
1. = (0.05): 8.d4 cxd4 9.cxd4 e4 10.Ne5 Bd6 11.Bc4 Be6 12.Bxd5 Bxd5 13.Nc3 Bxe5 14.dxe5 Bc4
2. = (0.00): 8.c4 Nf6 9.Nc3 Be7 10.Qb3 0-0 11.Bd2 Rb8 12.Nd5 Be6 13.Nxe7+ Qxe7 14.h3 Rbd8
(, 22.01.2007)
Of course, if you have seen the earlier moves, 8. c4 looks pretty illogical. (And yes, Rybka does eventually reject it after a bigger analysis - see below).
However, the move is also not that horrible objectively, probably second-best by a small margin. Rybka is a bit less dogmatic than other engines about "bad pawn structure" and this is intentional, I like it that way and think that it's appropriate.
Vas
New game - Rybka 2.2 32-bit, 120'/40+120'/40+120'
r1bqkb1r/pp3ppp/2n5/2pnp3/8/2PP1N2/PP2BPPP/RNBQ1RK1 w kq - 0 1
Analysis by Rybka 2.2 32-bit:
1. = (0.05): 8.d4 cxd4 9.cxd4 e4 10.Ne5 Bd6 11.Bc4 Be6 12.Bxd5 Bxd5 13.Nc3 Bxe5 14.dxe5 Bc4
2. = (0.00): 8.c4 Nf6 9.Nc3 Be7 10.Qb3 0-0 11.Bd2 Rb8 12.Nd5 Be6 13.Nxe7+ Qxe7 14.h3 Rbd8
(, 22.01.2007)
On my system, using the single cpu version (37kn/s), it took 11minutes 9 seconds to change it's mind to d4.
Vas, I also noticed that it likes doubled isolated pawns way more than it should. I think I have an example where it gladly creates 3 isolated doubled pawn islands at will and tries to defend the position, I'll post it soon. I duno, I find that kind of uncomforting (I guess you don't :P).
Vas, I also noticed that it likes doubled isolated pawns way more than it should. I think I have an example where it gladly creates 3 isolated doubled pawn islands at will and tries to defend the position, I'll post it soon. I duno, I find that kind of uncomforting (I guess you don't :P).
In positions like this, where Rybka's evaluation function doesn't have a strong preference, it will come down to "search luck". This can be affected by the hash size, hash contents, and MP non-determinism.
Re. doubled pawns, feel free to post any positions where you think Rybka is mis-behaving. Doubled pawns themselves are not that bad. What is bad is some of the things which sometimes come with doubled pawns. The main one there is a lack of pawn mobility, which sometimes hurts and sometimes doesn't.
For example, here is a doubled pawn you really don't want:
Vas
Re. doubled pawns, feel free to post any positions where you think Rybka is mis-behaving. Doubled pawns themselves are not that bad. What is bad is some of the things which sometimes come with doubled pawns. The main one there is a lack of pawn mobility, which sometimes hurts and sometimes doesn't.
For example, here is a doubled pawn you really don't want:
4k3/1pp2pp1/p1p4p/4P3/5PP1/7P/PPP5/4K3 w - - 0 1
Vas
Isn't it supposed to come to the same conclusion at the same depth (just different time required to reach it) regardless of mp or single????
Well, MP search is not totally deterministic. Sometimes a move will be favored at depth 13, other times not before depth 14 or 15. Basically, no two MP searches are exactly the same, as far as I would guess.
This also means that MP search is principally always different from single search. However, whatever the MP search will be like, the searches will usually come to the same conclusions in the same order, more or less.
This also means that MP search is principally always different from single search. However, whatever the MP search will be like, the searches will usually come to the same conclusions in the same order, more or less.
![[de]](/mwf/v12/flags/de.png "Germany")
After a couple of moves rybka still evaluates the position equal.
Bad square d4 against good square d5
Rybka 2.2n2 mp 32-bit 1 CPU - Fritz 10, rybka/F10, 180'/40+0'/0+30'
Analysis by Fritz 10:
µ (-1.07) 14...Dd7 15.Sg5 Sxg5 16.Lxg5 f5 17.Dh5 e4 18.f4 exf3 19.Dxf3 Se5 20.Dg2 Sxd3 00:00:02
Analysis by Rybka 2.2n2 mp 32-bit 1 CPU:
= (-0.08) 15.Ld2 b5 16.b3 Dd7 17.Kg2 Tfe8 18.Lc3 bxc4 19.bxc4 f6 20.Da4 Sed4 21.Sxd4 cxd4 00:00:12
= (-0.09) 15.a3 Dd7 16.Ld2 b5 17.Lc3 f6 18.b4 Scd4 00:00:25
= (-0.07) 15.Ld2 b5 16.b3 Dd7 17.Kg2 Tfe8 18.Lc3 bxc4 19.bxc4 f6 20.Da4 Sed4 21.Sxd4 cxd4 00:00:32
Bad square d4 against good square d5
Rybka 2.2n2 mp 32-bit 1 CPU - Fritz 10, rybka/F10, 180'/40+0'/0+30'
r2q1rk1/pp3ppp/2nbn3/2pNp3/2P5/3P1NP1/PP3P1P/R1BQRK2 b - - 0 1
Analysis by Fritz 10:
µ (-1.07) 14...Dd7 15.Sg5 Sxg5 16.Lxg5 f5 17.Dh5 e4 18.f4 exf3 19.Dxf3 Se5 20.Dg2 Sxd3 00:00:02
Analysis by Rybka 2.2n2 mp 32-bit 1 CPU:
= (-0.08) 15.Ld2 b5 16.b3 Dd7 17.Kg2 Tfe8 18.Lc3 bxc4 19.bxc4 f6 20.Da4 Sed4 21.Sxd4 cxd4 00:00:12
= (-0.09) 15.a3 Dd7 16.Ld2 b5 17.Lc3 f6 18.b4 Scd4 00:00:25
= (-0.07) 15.Ld2 b5 16.b3 Dd7 17.Kg2 Tfe8 18.Lc3 bxc4 19.bxc4 f6 20.Da4 Sed4 21.Sxd4 cxd4 00:00:32
There are always single positions where some preference looks bad.
On chessgames.com, Cecil Purdy (1st CC-Worldchampion) was quoted as saying: "By far the most important thing about weak pawns and weak squares is not to overrate their importance". I think Rybka´s style is generally pretty risky which makes her play attractive.
Jürgen
(P.S. Another quote by Purdy seems more popular - according to Wikipedia he died playing chess his final words being "I have a win, but it will take some time".)
On chessgames.com, Cecil Purdy (1st CC-Worldchampion) was quoted as saying: "By far the most important thing about weak pawns and weak squares is not to overrate their importance". I think Rybka´s style is generally pretty risky which makes her play attractive.
Jürgen
(P.S. Another quote by Purdy seems more popular - according to Wikipedia he died playing chess his final words being "I have a win, but it will take some time".)
Topic Rybka Support & Discussion / Rybka Discussion / Rybka plays an antipositional move, Vas can you explain?
Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill