Not logged inRybka Chess Community Forum
Up Topic Correspondence Chess / Correspondence Chess / Noob Q: the value of multiple engines for a "positional pos"
- - By PeterT Date 2012-05-18 19:37
I hope such a question has not already been addressed a lot in the past but I couldn't find anything in the forum so far...

In a correspondence game that I play against a very strong opponent, I have reached the following position after a really badly played opening. 
r1bq1rk1/1p2bppp/p1n1pn2/3p4/3P1B2/2NB4/PPPQNPPP/R4RK1 w - - 4 10

1. d4 Nf6 2. Bf4 d5 3. e3 c5 4. Nc3 cxd4 5. exd4 a6 6. Bd3 Nc6 7. Nge2 e6 8. Qd2 Be7 9. O-O O-O *

How to best approach such a position?

As I am out of book and probably even Nelson is, as this is clearly not a great approach by White,
I started to look up the theory in the opening books which have as the usual strategy an  attack
on the kingside. Not sure how to implement this given that the Nc3 is not well placed and d4 is a bit
endangered. Otherwise with regards to strategy there is not much I can think of or deduce from
looking at games with similar positions.
So I turned to engines... Having read a lot about what you all are saying on this forum I took some
of the recommended engines and created  the following table:

Name  Critter   Houdini   D. Shredder   Stockfish   Komodo   D. Rybka   Firebird
Version  1.4a      2.0c         12       2.1.1        3             4         12
PV          5         3          4           5        4             7          4
Ply        20        23          20         24       24            19         21
Nodes bln 1.0       4.0         1.4        0.8      2.0           2.0        1.0
10 h3   –0.11     –0.12                  –0.61                  –0.19      –0.12
10 a3   –0.19     –0.13       –0.08      –0.38    –0.19                    –0.15
10 Rad1 –0.19                                                   –0.20  
10 Nc1  –0.23                 -0.14      –0.52                  –0.22  
10 Rae1 –0.23                                     –0.18         –0.25  
10 Rfe1           –0.16                  -0.38    –0.19         –0.20      –0.22
10 Bg5                                   –0.41                  –0.22      –0.20
10 Rfb1                                  –0.56        
10 f3                         –0.20               –0.17         –0.23  
10 b3                         -0.19           

Is there any value in doing that? 
How to choose the best move now? 
Or would it have been better to use only 1 or 2 engines?
Maybe IDeA would have been a good approach but in my case it is not possible (I have no admin
rights on the computer I am using).
Parent - By Nelson Hernandez (Gold) Date 2012-05-18 20:37

> As I am out of book and probably even Nelson is

10.a3  KnightX 1.82-Frenzee 143, 0-1, 2005
10.h3  Rybka33, Rybka 3 Dynamic-Pacificrabbit, Rybka 3, 0-1, 2008.11.30 and also Antikavkaz, Rybka 3-Q9450 o-c, Rybka 3, 1-0, 2009.09.02
Parent - - By Uly (Gold) Date 2012-05-19 03:40 Edited 2012-05-19 03:51

> Is there any value in doing that?

Yes, though you seem to be wasting time by asking the engines for alternatives when probably they are not needed. And to use too many engines before they are needed. To minimize redundancy and so you can look deeper in the variations you'd do something like this:

Ask Critter for a move.

Critter h3 at -0.11.

You already know what move is its favorite one, you switch engine.

Houdini h3 at -0.12.

Redundancy, it means using these engines together may waste time, unless h3 is the best move or favored by most engines. Nothing to do here, switch engine.

Shredder a3 at -0.08

There, the second alternative move, now you have something to do: cross checking with Critter and Shredder on the position until they agree on what move is better. Since you are black, you want to reduce the score, so you take a look first at a3 and try to get Critter to like it.

You force a3 into the position and make Critter check it, now, please note that since you're a move away from the root, it's going to take Critter less time to check this move than from the root (a depth 19 here is equivalent of depth 20 on the root).

Critter [answer to a3] at -0.19

So it didn't like the move. Now, let's see what Shredder thinks about Critter's choice.

Shredder [answer to h3] ?

Oh, it seems you have this hole in your table. Okay, let's say Shredder gives it a -0.11.

Now, you want to analyze a3 with Critter and Shredder, to see if Shredder can convince Critter of a move change.

You play a3 and the answer Critter suggested into Shredder, and ask for a move, here, you want Critter to either give Shredder's line a score that is closer to zero and <-0.11 (Critter liking a3 better than h3) or refuting Shredder's line (at the time it gives a3 a score worse than -0.11, so it liked h3 better). Or the line reaches some depth without either happening (unclear line.)

So you give Critter the black side and Shredder the white side.

1.a3 x
Shredder [answer to x] 0.00

Shredder claims drawish position, this is better than -0.11 so you continue like this.

1.a3 x 2.y
Critter [answer to y] -0.17

And so on.

1.a3 [Shr/-0.08] a [Cri/-0.19] 2.b [Shr/0.00] c [Cri/-0.17] 3. d [Shr/-0.02] e [Cri/-0.24] 4. f [Shr/-0.14]

Here, Shredder shows a worse evaluation for this line than for h3.
You have also seen its idea on the line and where it leads to, why it doesn't work up to this depth, and a reason you'd want to avoid a3.

Now, you have 3 engines agreeing to h3 being best, you call another engine.

Stockfish Rfe1 at -0.38

Now, this is a very negative score. If Stockfish is right, then your better strategy on the position is to try to draw the game, and use Stockfish as your main engine to refute black's ideas towards 0.00 (by refuting Stockfish you're ensuring the opponent has nothing better to try). If Stockfish is wrong, then it has already given up and is not seeing white's defensive resources, you're better off dropping Stockfish from your line up.

So, how do you know if Stockfish is right or wrong? You crosscheck its ideas against current main engine Critter.

First, you see if Stockfish likes h3.

Stockfish [answer to h3] –0.61

And it's alarmingly low. Now, you ask Critter to score Stockfish's move.

Critter Rfe1 ?

Here's another hole in the table. Let's suppose it gives it a -0.19. Since you have a very low score for your main move, you want to bump it up to 0.00, so again, with Crosschecking, you give white to Critter, and black to Stockfish, and you try to refute its -0.61 for h3 to a score closer to 0 that is <-0.38, or to refute Critter's h3 to a score that is worse than -0.19:

1.h3 [Cri/-0.11] a [Sto/-0.61] 2. b [Crit/-0.16] c [Sto/-0.61] 3. d [Cri/-0.15] e [Sto/-0.51] 4. f [Cri/-0.20]

And there it is, Critter is giving a worse score to this line than to Stockfish'd Rfe1 move, it means from the root Critter was pruning this line. Stockfish being right for this line, it means the position may be worse than thought. You'd want to make Stockfish the new main engine, but since h3 has a higher score than Rfe1, you'd want to try to rescue it. So now, you ask for alternative moves for black in this line. While engines do this internally, and they do it from the tail backwards towards the root, I've found that better moves are likely to be found near the root, so you ask Critter for an alternative move to 2. b.

1.h3 a 2. g at -0.17

But since its score is worse than what 2. b was originally, you keep looking for alternatives, this time, to 3. d.

1.a3 a 2. b c 3. h at -0.21

There's none, and since after 3. d e, 4. f is only move with score -0.20, it means 2. b has score -0.20. The tree looks like this:

1.h3 a 2. b c 3. d e 4. f -0.20
              3. h        -0.21
       2. g               -0.17

You crosscheck to see if 2.g falls as well.

1.h3 [Cri/-0.11] a [Sto/-0.61] 2. g [Cri/-0.17] i [Sto/-0.66] 3. j [Cri/-0.13] k [Sto/-0.67] 4. l [Cri/-0.12] m [Sto/-0.56] 5. n [Cri/-0.14]

Okay, that's long enough, you can tag it unclear, at least Critter thinks that it's safe so far, but from this you have refuted 1.h3 to a -0.17, which means on a crosscheck with Stockfish as white and Critter as black, you only need to refute Critter's ideas to be closer to 0 and <-0.17. On Rfe1:

1.Rfe1 [Sto/-0.38] a [Cri/-0.19] 2.b [Sto/-0.46] c [Cri/-0.16]

And there it is! Please note Stockfish is refuting Critter from both sides in this position, which shows specific superiority here. In the process of this, Rfe1's Stockfish score fell to -0.46, so you cancel this line, go back to the root, and ask Stockfish for a move that isn't h3 or Rfe1.

Stockfish 1.Bg5 -0.41

Which, being the best score of the new main engine, becomes you main move, yet you have nothing on it yet. Here I would suggest analyzing Bg5 with Houdini (a hole in the table), and show the refuted line of h3 to it, to see if it also falls down to a score worse than 1.Bg5, or if Houdini finds a better white continuation that refutes Stockfish's black attack (what you did to Critter), until Bg5 falls or all engines agree Bg5 is best. You wouldn't introduce more engines until this is resolved.

This is about looking at these variations and understanding why they work or don't, eventually you'll see a variation of white that you like because black seems to have nothing to do against it, or one where white has fail highs on some variations so there's a chance your opponent blunders. As the refutation of Critter's h3 and its attack against Rfe1 show, this process is more effective than just letting Critter analyze the root position, or do interactive analysis with Critter alone, as it was missing Stockfish's moves.
Parent - By PeterT Date 2012-05-19 16:30
Wow, that is a nice lesson in correspondence chess. Thanks!
Up Topic Correspondence Chess / Correspondence Chess / Noob Q: the value of multiple engines for a "positional pos"

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill