Shokidoki wins UEC Cup!

[cdani]

Nice!! Congratulations!

[Evert]

Ah, I completely forgot to sign up for this! I'll try not to forget next year...

Either way: congratulations! Do you know whether the initial mini-Shogi is indeed a zugzwang position?

[Evert]

Congrats HG!
And Ferd too, it is a very close top-4!

We missed Nebiyu!

And Sjaak II. Funny enough Sjaak II is the most efficient engine in exposing Shokidoki's holes in the opening book. Perhaps it randomizes more.

Off the top of my head, it randomises move ordering for quiet moves in the first few plies of the PV (not just at the root). I'm not actually sure if that should result in more or less randomness in the end, but as an idea it seemed more reasonable than the more standard way of adding a random number to the evaluation at the root...

[hgm]

Ah, I completely forgot to sign up for this! I'll try not to forget next year...

Either way: congratulations! Do you know whether the initial mini-Shogi is indeed a zugzwang position?

I did finally find some lines for white (sente) that end in situations that even after 30 min of searching give a close-to-zero score. But the initial position does that too, so there is no guarantee that they are not lost after all.

And there is a crucial line that makes makes the break-out of white from a symmetric position possible. In the following position

Code: Select all

r . . . k
. s b g p
. . . . .
P G B S .
K . . R .

white can initiate the break-out through Bb1, Sc1, Sc2, after which it can move safely to the third rank on either b3 or d3. But superficially this seems to be refuted by

1. Bb1 Sb3?! 2. Sc3! Sxb2 3. Sxb2 G@c2 4. Bxb2 Rxa2 5. Kb1 Rxb2 6. Kxb2 S@c3 7. Ka3 Sxc2 8. Ra1

which gains black a Pawn and totally destroys the white King safety. But funny enough this position seems lost for black, although the situation is extremely murky. So white can safely play 1. Bb1 after all.

[Ferdy]

Congrats HG!
And Ferd too, it is a very close top-4!

Thanks. There is upcoming GAT Cup to be held early next year prepare your engine we might get invited.

http://minerva.cs.uec.ac.jp/~uec55shogi ... 2%E7%B2%F1

[Ferdy]

Ah, I completely forgot to sign up for this! I'll try not to forget next year...

If we get invited we could probably play again. Check the GAT Cup, early next year.

http://minerva.cs.uec.ac.jp/~uec55shogi ... 2%E7%B2%F1

[hgm]

Indeed, I saw that announcement appear too, yesterday. I doubt many of us will be invited, however; it seems a pretty small event with only 4 rounds of Swiss. I am not sure what the goal of this meeting is; if it is to stimulate development of Japanese programs they are probably not going to invite many programs from abroad. OTOH, if the purpose is to have as strong a field as possible, they should certainly make sure the top-4 of the 9th UEC Cup participate.

I am not sure where Sjaak II stands relative to TJshogi5x5. From UEC 9 it seems there is a clear gap both above and below TJshogi. Not only in the result points, but also in the matrix: TJshogi lost all games against the programs finishing above it, but won from all programs finishing below it, and now other upsets occurred in the high vs low groups. So rating programs would have no way to estimate the gap between the TJshogi and all the others, based on the tournament results.

[Evert]

I am not sure where Sjaak II stands relative to TJshogi5x5. From UEC 9 it seems there is a clear gap both above and below TJshogi. Not only in the result points, but also in the matrix: TJshogi lost all games against the programs finishing above it, but won from all programs finishing below it, and now other upsets occurred in the high vs low groups. So rating programs would have no way to estimate the gap between the TJshogi and all the others, based on the tournament results.

In UEC 8 they scored equal, but it isn't clear how representative this is for what would have happened in UEC 9 if SjaakII had played there. TJshogi did win its direct game against SjaakII...

[hgm]

Yes, but I think you made progress between 8th UEC (where Sjaak II was still in a beta stage, IIRC) and the Olympiad in Leiden, while TJshogi has been version 0.19 for years. (While the version on Tony's website still is 0.10.)

And Sjaak II did beat both Rigan and Shokidoki, in the 8th UEC Cup!

[hgm]

I ran a match of 20+0 ponder-on games between Sjaak II 1.1.1 and TJshogi5x5 0.19 (which Tony confirmed is still the latest version). Sjaak II was leading 52-31 when I stopped it. There were 3 games that hung by Sjaak II playing a null move; in 2 of the 3 Sjaak was just a few moves away from checkmating TJ, in the 3rd it had a negative score, and TJ was even pondering about a mate. (I am sure many of the games were duplicats; I think TJ plays completely reproducibly, and I did not use any openings or start positions.)

So it seems Sjaak II is some 100 Elo stronger than TJ, currently.