Battle of the Goths 2012 (live broadcast)

[hgm]

After 3 cycles in the playoffs, the cross table is:

Code: Select all

                              Biha Spar TJch Nebi Joke SMIR tscp
 1. Bihasa v                  #### 1111 1111 1111 1=01 =111 1111
                              #### 11   11   11   11   11   11     94%  34.0 (588.0, 560.0)

 2. Spartacus                 0000 #### 1011 10=1 0101 1101 1111
                              00   #### 10   1110 11   11   1111   66%  26.5 (682.0, 344.8)

 3. TJchess                   0000 0100 #### 1011 1111 1111 1111
                              00   01   #### 10   01   10   11     61%  22.0 (660.0, 286.0)

 4. NebiyuChess               0000 01=0 0100 #### 1101 0111 0011
                              00   0001 01   #### 01   11   1111   49%  19.5 (738.0, 249.8)

 5. Joker                     0=10 1010 0000 0010 #### 1111 0111
                              00   00   10   10   #### 01   01     43%  15.5 (699.0, 232.5)

 6. SMIRF MS                  =000 0010 0000 1000 0000 #### 0111
                              00   00   01   00   10   #### 11     26%   9.5 (735.0, 125.5)

 7. tscpgothic                0000 0000 0000 1100 1000 1000 ####
                              00   0000 00   0000 10   00   ####   12%   5.0 (854.0,  79.5)

In the third cycle, Bihasa had a clean sweep: 12 out of 12! The gap between Bihasa and the number 2 Spartacus has now grown so large (7.5 points, while it has 4 games less!) that it can be considered unbridgeable. The same could be true for the gap between Joker80 and Smirf. With a bit of luck, TJchess10x8 could still catch up with Spartacus, and Joker80 with Nebiyu. (Because they have fewer games, as the games of Spartacus and Nebiyu still standing from the Qualifier are already incorporated in the above cross table!)

The final cycle, from the Carrera opening position, has now started. Let's hope for an exciting finish!

[George Tsavdaris]

Code: Select all

    name                       perc.  pnt games  Buchh    S-B

 1. Bihasa v                    96%  45.0 / 47  (916.5, 876.5)
 2. Spartacus                   59%  27.5 / 47  (1039.0, 453.3)
 3. TJchess                     57%  26.0 / 46  (1068.0, 424.0)
 4. NebiyuChess                 51%  23.5 / 46  (1072.0, 388.8)
 5. Joker                       48%  22.0 / 46  (1081.0, 396.0)
 6. SMIRF MS                    30%  14.0 / 46  (1140.0, 222.5)
 7. tscpgothic                  10%   5.0 / 48  (1264.0, 105.0)

bihasa has 45 out of 47 with 44 wins 2 draws and just one lose.
One could say unbelievable!
But not me. I say under-performance lol.

I have predicted in the start that Bihasa would finish with 47.5 out of 48 and i was too close, but watching the games i've seen that Bihasa had been endangered to lose many more games. One thing is certain. Programs play awfully the opening in 10x8 Chess. They literally suck!

And that is the reason for Bihasa's danger of losing more games. It played the opening terribly so in 10x8 Chess you can pay it much more easily than Chess i think, because the King in the middle is instant death. Of course its opponents didn't manage to convert at the end and got outsmarted.

Great tournament and since Bihasa will win as it seems its final game against Spartacus(the final game for it also), TJChess will have the opportunity with 2 wins out of 2 games to go for the 2nd place.

[Evert]

I have predicted in the start that Bihasa would finish with 47.5 out of 48 and i was too close, but watching the games i've seen that Bihasa had been endangered to lose many more games. One thing is certain. Programs play awfully the opening in 10x8 Chess. They literally suck!

Isn't this the same for normal chess if you let engines play without opening book?
It doesn't really help of course if you don't play a variant yourself and aren't quite sure what the opening goals should be (or indeed, knowing what to put in an opening book if you were to make one). In Sjaak I have an evaluation component that disappears as pieces are developed that tries to encourage development of minor pieces, centralisation and castling. No idea how much this actually helps in practice though.

I would be very interested in seeing some more discussion/suggestions for improving engine opening play (without an opening book), but it's hard to find anything.

[Ferdy]

I have predicted in the start that Bihasa would finish with 47.5 out of 48 and i was too close, but watching the games i've seen that Bihasa had been endangered to lose many more games. One thing is certain. Programs play awfully the opening in 10x8 Chess. They literally suck!

And that is the reason for Bihasa's danger of losing more games. It played the opening terribly so in 10x8 Chess you can pay it much more easily than Chess i think, because the King in the middle is instant death. Of course its opponents didn't manage to convert at the end and got outsmarted.

It is only in v3.2 that I added more penalty for undeveloped knights and bishops (specifically addressing the opening problem),
seeing previous versions doing solo piece attack . I guess v3.2 does not lose or draws.

[Ferdy]

I have predicted in the start that Bihasa would finish with 47.5 out of 48 and i was too close, but watching the games i've seen that Bihasa had been endangered to lose many more games. One thing is certain. Programs play awfully the opening in 10x8 Chess. They literally suck!

Isn't this the same for normal chess if you let engines play without opening book?
It doesn't really help of course if you don't play a variant yourself and aren't quite sure what the opening goals should be (or indeed, knowing what to put in an opening book if you were to make one). In Sjaak I have an evaluation component that disappears as pieces are developed that tries to encourage development of minor pieces, centralisation and castling. No idea how much this actually helps in practice though.

I would be very interested in seeing some more discussion/suggestions for improving engine opening play (without an opening book), but it's hard to find anything.

I believe after the minor pieces are developed and some pawn pushes to control couple of squares in the center and after castling - opening is done.

[hgm]

Final standings of the playoffs:

Code: Select all

Cross table, sorted by score percentage, Buchholz, SB

                              Biha Spar TJch Nebi Joke SMIR tscp
 1. Bihasa v                  #### 1111 1111 1111 1=01 =111 1111
                              #### 1111 1111 1111 1111 1111 1111   96%  46.0 (976.0, 934.3)

 2. Spartacus                 0000 #### 1011 10=1 0101 1101 1111
                              0000 #### 1000 1110 1101 1100 1111   57%  27.5 (1124.0, 472.3)

 3. TJchess                   0000 0100 #### 1011 1111 1111 1111
                              0000 0111 #### 1000 0100 101= 1111   57%  27.5 (1124.0, 457.3)

 4. NebiyuChess               0000 01=0 0100 #### 1101 0111 0011
                              0000 0001 0111 #### 0101 1111 1111   51%  24.5 (1148.0, 425.3)

 5. Joker                     0=10 1010 0000 0010 #### 1111 0111
                              0000 0010 1011 1010 #### 011= 0111   48%  23.0 (1160.0, 431.8)

 6. SMIRF MS                  =000 0010 0000 1000 0000 #### 0111
                              0000 0011 010= 0000 100= #### 1111   30%  14.5 (1228.0, 240.8)

 7. tscpgothic                0000 0000 0000 1100 1000 1000 ####
                              0000 0000 0000 0000 1000 0000 ####   10%   5.0 (1304.0, 109.5)

Bihasa has won this tourney in the most convincing way! The battle for second place was thrilling: Spartacus seemed to have a comfortable lead on TJchess10x8, but in the last cycle lost twice against outsider SMIRF (always dangerous, and one of the two engines that scored against all opponents!) In addition it lost twice to TJ, and this was enough for TJ to grab second place in the last round, if it would have defeated SMIRF twice. But in the very last game of the tourney, SMIRF managed to draw by a perpetual from a very bad position, thus creating a tie between Spartacus and TJ. Sonnenborn-Berger points decide this tie in favor of Spartacus.

I have now started a multi-gauntlet with Heretic 0.3 and Sjaak 470, the fixed versions of the engines for which the fix came too late to qualify for the playoffs. They will unofficially play against the playoff participants.

[smrf]

Thank you for performing this 10x8 tournament!

Next year maybe the number of participants is doubling again.

[mar]

Thank you for the tourney hgm and congratulations to the World champion Ferdy Mosca and Bihasa! I hope the Shannon trophy is on its way already.

Martin

[George Tsavdaris]

Isn't this the same for normal chess if you let engines play without opening book?

No. Todays Chess engines of course without opening book they play much much worse the opening, but they play very sensible and good moves overall. Not all times of course.
10x8 CRC engines on the other hand, play terribly.
And i mean terribly.

I would be very interested in seeing some more discussion/suggestions for improving engine opening play (without an opening book), but it's hard to find anything.

•Play Pawn moves first, that open the long diagonals for Bishops.
•Develop your Knights before you develop your Archbishop/Chancellor and Queen.
•Create a skeleton of Pawns before moving the Chancellor and Archbishop and the Queen.
•NEVER(ok almost never) develop Chancellor or Archbishop in the 3rd(6th for black) rank in the opening. Usually it's bad. 2nd(7th) row is always the best. (of course that doesn't mean the first move of A or C should not be on 3rd rank. No. It means that in the opening phase they should not be placed there. But if the position is more advanced and the characteristics of it, i.e the skeleton of Pawns is well defined and the other pieces have been developed, then if a good square for A or C is in the 3rd rank, then yes you should place it there of course)

•Be aggressive on the center and don't let the opponent to dominate it.
•Bishop's development should not be rushed. They can wait.

The general rule of course is avoid playing a piece twice in the opening.

Note of course, that if the opponent plays badly in the opening and you can gain something from that, by breaking the above rules, you should do it by all means and break the rules.

[Evert]

No. Todays Chess engines of course without opening book they play much much worse the opening, but they play very sensible and good moves overall. Not all times of course.

Is this because they tend to search deep enough to reach a middle-game position, where they have some idea for whether pieces are well-placed or not?
Or is there some other reason?
For chess variants the evaluation is not as good as for normal chess engines, so that certainly doesn't help.

•Play Pawn moves first, that open the long diagonals for Bishops.
•Develop your Knights before you develop your Archbishop/Chancellor and Queen.
•Create a skeleton of Pawns before moving the Chancellor and Archbishop and the Queen.
•NEVER(ok almost never) develop Chancellor or Archbishop in the 3rd(6th for black) rank in the opening. Usually it's bad. 2nd(7th) row is always the best. (of course that doesn't mean the first move of A or C should not be on 3rd rank. No. It means that in the opening phase they should not be placed there. But if the position is more advanced and the characteristics of it, i.e the skeleton of Pawns is well defined and the other pieces have been developed, then if a good square for A or C is in the 3rd rank, then yes you should place it there of course)

•Be aggressive on the center and don't let the opponent to dominate it.
•Bishop's development should not be rushed. They can wait.

The general rule of course is avoid playing a piece twice in the opening.

That's all very hard to translate into an algorithm.
I guess what's needed is a combination of two things: evaluation and move ordering. I notice that if I just put things like "develop your knight before your queen", then the program will happily develop its queen first because it thinks it can develop its knight afterwards anyway and then it's back to the situation where both have been developed.

In Sjaak, I currently give a penalty for pieces that have not been developed, a penalty for not having castled (but being able to) and a larger penalty for not having castled and no longer being able to. This at least gets it to move its pieces out and stick its king in the corner.

I can try the following: in the opening, sort pawn moves to the centre higher up in the move list, they're probably good. Sort development moves of minor pieces higher than those of heavy pieces. When developing heavy pieces, favour moves that keep them "behind the lines" (so don't place them before the pawns). Perhaps also increase the weight of the centre in the evaluation until most pieces have been developed to encourage pieces to control the centre.
Does this sound like a reasonable thing to try?