can chess program see win for white after e4 h6 d4 a5?

[Brunetti]

For example, currently FB1.2 against itself +2/-0/=8 for white at 3min/game. Even with 10 games sufficient to tell that it will be less than 80%.

Shouldn't all games played against itself from the same position be identical? How do you explain the different results?

If I had to test this position, I would run a standard double round robin tournament with, say, 20 different engines/engine versions in the same rating range, and then compare White's score against the normal 54 or 55%. Possibly Harun could do that in a while

Alex

[Uri Blass]

For example, currently FB1.2 against itself +2/-0/=8 for white at 3min/game. Even with 10 games sufficient to tell that it will be less than 80%.

Shouldn't all games played against itself from the same position be identical? How do you explain the different results?

If I had to test this position, I would run a standard double round robin tournament with, say, 20 different engines/engine versions in the same rating range, and then compare White's score against the normal 54 or 55%. Possibly Harun could do that in a while

Alex

I agree

Nothing wrong with playing the same engine against itself but it should be no more than one game(unless you use different time controls).

I think that the sweningen system can be used when Firebird White,Rybka white,Naum White,Stockfish White can play against the team of the same program with black.

Every White program is playing 4 games including game against the same engine so we get 16 games.

Edit:I can explain different results because engines are not deterministic with 4 processors(and even with 1 processor time may be also slightly different and cause the program to search slightly more nodes in one of the games that can cause a different move) but the results are clearly related and we get less information from related results.

Uri

[bob]

and winning...

You just never know what might happen on a given day:

[Event "Jason"]
[Site "?"]
[Date "2007.11.09"]
[Round "8.5"]
[White "Fruit 05/11/03"]
[Black "Deep Shredder 11 w32 1CPU"]
[Result "1-0"]
[ECO "A00"]

1.h3 e5 2.e3 d5 3.d4 Nd7 4.Nf3 e4 5.Nfd2 Ngf6 6.c4 c6 7.Nc3 Bd6 8.Be2 Qe7
9.O-O O-O 10.cxd5 cxd5 11.Nb5 Nb8 12.Nxd6 Qxd6 13.Nb1 a6 14.a4 Bf5 15.b3
Rc8 16.Ba3 Qb6 17.Qd2 Qe6 18.Nc3 Nc6 19.Rfc1 h6 20.b4 a5 21.bxa5 Nxa5 22.
Qa2 Qd7 23.Rab1 Be6 24.Rb6 Rc7 25.Bc5 Qc8 26.Nb5 Rc6 27.Nd6 Qc7 28.Rxc6
bxc6 29.Bb4 Qd7 30.Rb1 Ne8 31.Bxa5 Nxd6 32.Bb4 Nc4 33.a5 Qc7 34.Bc3 Kh7
35.Bxc4 dxc4 36.Rb6 f5 37.Qb1 Qd6 38.Qb4 Qxb4 39.Bxb4 Bd7 40.Kf1 Rc8 41.
Rb7 Be6 42.a6 Ra8 43.a7 c3 44.Ke1 Bc8 45.Rc7 Be6 46.Bxc3 Bd5 47.Bb4 Be6
48.Bc5 Bc8 49.Bf8 Kg6 50.Rxg7+ Kf6 51.Rc7 Ke6 52.Rxc6+ Kd5 53.Rc7 h5 54.
Bh6 Ke6 55.Bf4 Kd5 56.Rc5+ Ke6 57.d5+ 1-0

We even have humans punishing computers with the *cough* powerful 1.h3:

[Event "ICC 5 3 04/16/97"]
[Site "Internet Chess Club INT"]
[Date "????.??.??"]
[Round "?"]
[White "insight"]
[Black "Crafty"]
[Result "1-0"]
[ECO "A00d"]
[Variation "Clemenz Opening"]

1.h3 e5 2.h4 Be7 3.d4 exd4 4.Nf3 c5 5.c3 dxc3 6.Nxc3 Nf6 7.e4 O-O 8.e5 Ne8
9.Bd3 Nc6 10.Bxh7+ Kh8 11.Bb1 d6 12.Ng5 g6 13.h5 Bxg5 14.hxg6+ Kg8 15.Rh8+
Kg7 16.Rh7+ Kg8 17.Qh5 Qf6 18.exf6 Bd2+ 1-0

in 1997 at under 50K nodes per second. Might be a bit harder today where search depth has more than doubled.

[Dann Corbit]

Here is the result after 42 games:

Code: Select all

Games        :     42 (finished)

White Wins   :     21 (50.0 %)
Black Wins   :      2 ( 4.8 %)
Draws        :     19 (45.2 %)
Unfinished   :      0

White Perf.  : 72.6 %
Black Perf.  : 27.4 %

About 170 Elo difference.

[Dann Corbit]

Here is the result after 42 games:

Code: Select all

Games        :     42 (finished)

White Wins   :     21 (50.0 %)
Black Wins   :      2 ( 4.8 %)
Draws        :     19 (45.2 %)
Unfinished   :      0

White Perf.  : 72.6 %
Black Perf.  : 27.4 %

About 170 Elo difference.

However, when I throw out all duplicates (because it was engine verses itself), I get this:

Code: Select all

c:\pgn\step1>type general.dat

Games        :     25 (finished)

White Wins   :      4 (16.0 %)
Black Wins   :      2 ( 8.0 %)
Draws        :     19 (76.0 %)
Unfinished   :      0

White Perf.  : 54.0 %
Black Perf.  : 46.0 %

ECO A =      0 Games ( 0.0 %)
ECO B =      0 Games ( 0.0 %)
ECO C =      0 Games ( 0.0 %)
ECO D =      0 Games ( 0.0 %)
ECO E =      1 Games ( 4.0 %)


  stoblu.pgn (      7): [Result "0-1"]
  stoblu.pgn (     41): [Result "1/2-1/2"]
  stoblu.pgn (     76): [Result "1/2-1/2"]
  stoblu.pgn (    111): [Result "1/2-1/2"]
  stoblu.pgn (    144): [Result "1/2-1/2"]
  stoblu.pgn (    173): [Result "1-0"]
  stoblu.pgn (    207): [Result "0-1"]
  stoblu.pgn (    238): [Result "1/2-1/2"]
  stoblu.pgn (    272): [Result "1/2-1/2"]
  stoblu.pgn (    313): [Result "1/2-1/2"]
  stoblu.pgn (    346): [Result "1-0"]
  stoblu.pgn (    376): [Result "1/2-1/2"]
  stoblu.pgn (    401): [Result "1/2-1/2"]
  stoblu.pgn (    434): [Result "1/2-1/2"]
  stoblu.pgn (    463): [Result "1/2-1/2"]
  stoblu.pgn (    497): [Result "1/2-1/2"]
  stoblu.pgn (    527): [Result "1/2-1/2"]
  stoblu.pgn (    558): [Result "1/2-1/2"]
  stoblu.pgn (    591): [Result "1/2-1/2"]
  stoblu.pgn (    619): [Result "1/2-1/2"]
  stoblu.pgn (    648): [Result "1/2-1/2"]
  stoblu.pgn (    680): [Result "1-0"]
  stoblu.pgn (    713): [Result "1/2-1/2"]
  stoblu.pgn (    745): [Result "1/2-1/2"]
  stoblu.pgn (    776): [Result "1-0"]

[muxecoid]

If black is heavily booked up for this opening while white uses vanilla book I guess the chances are equal even on medium-long time controls.

It will be hard to create such opening book as it will require human supervision to promote the moves that utilize black's control over b4 and g5. I guess the idea is to castle opposite sides while keeping the center somewhat closed.

[Karlo Bala]

I wonder what is your opinion:

1)Is it correct that white objectively has a winning position after these 2 mistakes by black

White has a big advantage. Between peer engines, white should win at least 90% I think. If white is me and black is Rybka, then that is another story.

2)Can chess program see a winning score based on deep search(the important thing is not a big score for white but seeing significant increase in score)?

Big scores in the opening are almost always a mistake, unless there is a whole lot of wood missing and the search is deep. I never trust a deep score in the opening with a search that is under one hour on a fast machine.

3)Does white always win in comp-comp games at long time control between top programs(Naum4.2 Stockfish1.6.3,Rybka3,Firebird1.2)?

I guess almost always.

4)What is the best plan for white?

Nf3 is my current guess, but I am not sure.

White indeed has big advantage. 3 tempi in opening worth about a pawn, and white has almost full 3 tempi more. However, playing both Nc3 and Nf3 at same time, is terribly wrong because position stay closed. White should play for big center and try to open position at right moment.

[S.Taylor]

I wonder what is your opinion:

1)Is it correct that white objectively has a winning position after these 2 mistakes by black

2)Can chess program see a winning score based on deep search(the important thing is not a big score for white but seeing significant increase in score)?

3)Does white always win in comp-comp games at long time control between top programs(Naum4.2 Stockfish1.6.3,Rybka3,Firebird1.2)?

4)What is the best plan for white?

Is it Nc3 and Nf3 or maybe it should include at least one of c4 and f4 that fide master suggested in discussion in the following thread

http://chesstempo.com/chess-forum/gener ... 999.0.html

Some analysis by Rybka3 does not show a significant increase in score but of course more time is needed and if somebody with good hardware can give chess programs 24 hours to analyze then it may be interesting.

spinnakr2001 - signofjonas, Yahoo! Chess Game 2010
[D]rnbqkbnr/1pppppp1/7p/p7/3PP3/8/PPP2PPP/RNBQKBNR w KQkq a6 0 1

Analysis by Rybka 3 1-cpu 32-bit :

3.Nb1-c3
² (0.48) Depth: 2 00:00:00
3.Nb1-c3
² (0.37) Depth: 3 00:00:00
3.Nb1-c3
² (0.37) Depth: 4 00:00:00
3.Nb1-c3 d7-d6
² (0.42) Depth: 5 00:00:00
3.Nb1-c3 d7-d6 4.Ng1-f3
² (0.46) Depth: 6 00:00:00 3kN
3.Nb1-c3 d7-d6 4.Ng1-f3 Ng8-f6
² (0.48) Depth: 7 00:00:00 4kN
3.Nb1-c3 d7-d6 4.Ng1-f3 Ng8-f6 5.Bf1-d3
² (0.46) Depth: 8 00:00:00 9kN
3.Nb1-c3 d7-d6 4.Ng1-f3 Ng8-f6 5.Bf1-d3 g7-g5
² (0.57) Depth: 9 00:00:01 20kN
3.Nb1-c3 d7-d6 4.Ng1-f3 g7-g5 5.Bf1-c4 g5-g4 6.Nf3-h4
² (0.50) Depth: 10 00:00:01 38kN
3.Nb1-c3 d7-d6 4.Ng1-f3 g7-g5 5.Bf1-c4 g5-g4 6.Nf3-h4 Ng8-f6
² (0.55) Depth: 11 00:00:03 76kN
3.Nb1-c3 d7-d6 4.Ng1-f3 g7-g5 5.Bf1-c4 g5-g4 6.Nf3-h4 Nb8-c6 7.Bc1-e3 h6-h5
² (0.57) Depth: 12 00:00:07 162kN
3.Nb1-c3 d7-d6 4.Ng1-f3 g7-g5 5.h2-h3 Bf8-g7 6.Bf1-e2 c7-c6 7.0-0 Ng8-f6 8.e4-e5
² (0.61) Depth: 13 00:00:14 307kN
3.Nb1-c3 d7-d6 4.Ng1-f3 g7-g5 5.h2-h3 Bf8-g7 6.Bf1-e2 c7-c6 7.0-0 Ng8-f6 8.e4-e5
² (0.61) Depth: 14 00:00:25 544kN
3.Nb1-c3 d7-d6 4.Ng1-f3 g7-g5 5.h2-h3 Bf8-g7 6.Bf1-e2 c7-c6 7.0-0 Ng8-f6 8.e4-e5 Nf6-d5
² (0.60) Depth: 15 00:00:50 1114kN
3.Nb1-c3 d7-d6 4.Ng1-f3 g7-g5 5.h2-h3 Bf8-g7 6.Bf1-e2 c7-c6 7.0-0 Ng8-f6 8.e4-e5 Nf6-d5 9.e5xd6 Qd8xd6 10.Nc3-e4
² (0.61) Depth: 16 00:02:03 2625kN
3.Nb1-c3 d7-d6 4.Ng1-f3 g7-g5 5.h2-h3 Bf8-g7 6.Bf1-e2 c7-c6 7.0-0 Ng8-f6 8.e4-e5 Nf6-d5 9.e5xd6 Qd8xd6 10.Nc3-e4 Qd6-g6
² (0.61) Depth: 17 00:03:42 4831kN
3.Nb1-c3 d7-d6 4.Ng1-f3 g7-g5 5.h2-h3 Bf8-g7 6.Bf1-d3 Ng8-f6 7.0-0 Nb8-c6 8.Rf1-e1 0-0 9.a2-a3 e7-e6 10.Bc1-e3
² (0.59) Depth: 18 00:08:49 11716kN
3.Nb1-c3 d7-d6 4.Ng1-f3 g7-g5 5.h2-h3 Bf8-g7 6.Bf1-c4 Nb8-d7 7.0-0 e7-e6 8.Bc1-e3 Ng8-e7 9.Bc4-d3 c7-c6 10.Qd1-d2 0-0
² (0.61) Depth: 19 00:23:37 31017kN

(, 30.03.2010)

This kind of question has probably been around for many years, and we can only ask it again every few years, when the new generations of computers might have reached the ability to teach us wonderful new things.
I've recently been reading a very special little book by Edmar Mednis "How to play good opening moves", and he pays attention to every possible opening first move for white and black, and those which he considers "weak", he explains why. If the computer understands everything he had in mind, then who knows? perhaps it IS a win for white with perfect play both sides!

[S.Taylor]

If the best and fastest programs do not go up and up when left to evaluate, then it looks like the plus score after 20-30 plies is for the same reason it was a plus at the first second, which seems to be the general positional opening progress, and not that something subtantial, or further weakness has come from it.

Perhaps the actual computer brain is not yet powerful enough to know how to take advanage of many weak moves and positions. The computer still has to rely alot on what it is taught, and this is great, but is still a little bit like a toy, which cannot innovate or teach, as it in itself does not undestand, as it has neither been taught everything, nor can it see far enough from calculaton or independent intuition.

[Milos]

Shouldn't all games played against itself from the same position be identical? How do you explain the different results?

The easiest solution is to just throw away all the duplicates. There should not be many duplicates anyway, because engines are not working deterministically.
First you have the impact of OS (schedulers and other processes running) which introduces a lot of noise in engine thinking process. Secondly, even if you had a dedicated OS which would give all of the processor time to an engine, multithreaded engines are again not deterministic by definition (for the reasons of how multicore processors work). There is a lot of noise and a game consists in quite a lot of moves. Chance to have divergence somewhere is increasing with each new move played.