Page 1 of 3

Longest games

Posted: Sat Mar 23, 2019 5:42 pm
by Henk
yawn

Re: Longest games

Posted: Sat Apr 06, 2019 5:39 pm
by Henk
Hi hi hi hi Fairy max unable to win KNBk endgame.

"
Shame shame shame hey shame on you
If you can't dance too
I said shame shame shame shame shame shame shame
Shame on you
If you can't dance too
"


Re: Longest games

Posted: Sun Apr 07, 2019 4:45 am
by MikeB
I get least at one long game like this around every 400 games or so. All my games are always played to the bitter end.


Re: Longest games

Posted: Sun Apr 07, 2019 5:10 am
by MikeB
here's one of the longer decisive games - it looks like 66. Rh7 was the losing move - after that move McCain gets a big score jump - from +.62 to +3.63 for black. Asmfish must have a bug , if you set up the position , it never considers 66.Rh7 , but it plays that move in 30 sec game with 2.5 second increment


Re: Longest games

Posted: Sun Apr 07, 2019 6:30 am
by Modern Times
If you play Lc0 with all adjudication turned off (which I do), you get excruciatingly long games.

Re: Longest games

Posted: Sun Apr 07, 2019 6:38 am
by hgm
Henk wrote:
Sat Apr 06, 2019 5:39 pm
Hi hi hi hi Fairy max unable to win KNBk endgame.
Actually Fairy-Max 5.0 should be able to do that. In end-games like KBNK it is essential the static evaluation knows which corner to drive the King to. Otherwise it will trap him in the corner where no mate can be forced, and never allow him to escape.

But because Fairy-Max is a general chess-variant engine, it cannot have such variant-specific knowledge. Whatever it does, it must also work when it has to checkmate a bare King, say, with a Phoenix and a Mortar in Team-Mate Chess.

So I implemented the following solution: apart from the normal centralization table, which would push the bare King into any corner, there are also tables that only push the bare King in diametrically opposite corners, and keep it out of the other two. And whenever a bare King reaches a corner, this will switch the table it uses to the one that penalizes it in the other set of corners. Now usually the search depth of Fairy-Max is large enough to see the checkmate when the King is already cornered, in which case the static evaluation doesn't affect the search, and the switch of PST remains unnoticed. But if the bare King is in the wrong corner, there will be no mate in view, and the static evaluation will now promote driving the King to the nearest other corner, along an edge. Where hopefully the mate will be possible, and then spotted. This usually works well enough to win KBNK (as well as KMXK).

Re: Longest games

Posted: Sun Apr 07, 2019 6:43 am
by Modern Times
Check out the insanity of this game where it takes Lc0 almost 300 moves to checkmate:


Re: Longest games

Posted: Mon Apr 08, 2019 8:58 am
by hgm
Well, what can you expect? It is explicitly trained for not being able to make progress. If at some stage it would prefer certain wins that take a few moves over certain wins that take many, many moves it is considered an error, and the NN is adjusted to assign the score +1.00 to both certain wins.

Re: Longest games

Posted: Mon Apr 08, 2019 9:17 am
by Daniel Anulliero
Henk wrote:
Sat Apr 06, 2019 5:39 pm
Hi hi hi hi Fairy max unable to win KNBk
Hihihi SchaakMachine unable To win a single game
Shaaame on you 😁
Does It know KKQ or KKR ? Hooope sooo 😉
Shame shame shame.....

Re: Longest games

Posted: Mon Apr 08, 2019 11:04 am
by Henk
Yes my engine is still in a bad state. Only once in a while able to win against fairy max.
KQk no problem. Don't know about KRk. Of course my engine can't win KBNk. Much too difficult.
But I was surprised that fairy max wasn't able to finish KBNk for it usually plays endgame very well.