I think nothing unusual on this game, just that Lc0 is again a pawn down but have a lot of pc activity for 1 pawn, and k activity in endgame.
Arasan probably saw a forced mate with or without giving up both his rooks.
The win for Lc0 is smooth, avoiding desperado rook tricks for a stalemate.
ernest wrote: ↑Fri Sep 07, 2018 12:14 am
Lc0 is unable to win the obvious endgame, and suddenly (Black moves 144 and 145) Arasan stops fighting, and surrenders its 2 rooks !...
Yeah, I don't see any big error in the endgame. Black is just lost.
I was actually watching the first part of this game, and I think Arasan went wrong early in the opening. 8. .. f6 looks like an error. Arasan then expanded on the kingside with .. g5 and .. h5, but this was futile because White wound up castling queenside.
MikeGL wrote: ↑Fri Sep 07, 2018 8:17 am
Arasan probably saw a forced mate with or without giving up both his rooks.
The win for Lc0 is smooth, avoiding desperado rook tricks for a stalemate.
Hi Mike,
For Arasan, you may be right, I will ask Jon...
But I disagree when you say "the win for Lc0 is smooth", for about 40 moves Lc0 has no idea about how to win, eval is not advancing a bit, it's a dismal show (especially with Lc0 supposedly using Sygyzy). Comparing with the moves proposed by alpha-beta programs (e.g. Stockfish), you can see the big failings.
Maybe somebody from the Lc0 team can explain more.
lc0 is not using tb in 4C.
The version playing is capable of only using WDL tables, but that makes the late endgame play super weird. The plan is to use both DTZ and WDL together once the code has been properly tested. In the next tournament, I guess, as changes to the engines are not allowed during or between rounds.
The endgame against arasan looked hysterical. At one time I thought lc0 was demonstrating how many ways you can arrange two rooks on an open a-file while not making any progress.
The endgame against Texel in a later game was impressive - rooks and pawns against bishop, knight and pawn. Curious.
ernest wrote: ↑Fri Sep 07, 2018 8:44 pm
Indeed, Black is lost, but why does Arasan choose such a short death ?
(especially after resisting so well to the previous long series of Lc0's aimless moves...)
If you mean 144 .. Rd7, after a few seconds Arasan is getting a "TbLoss" score, indicating a certain loss (it starts hitting the 6-man tbs).
Arasan could have played so as to delay the mate further. But with Syzygy bases, you don't get a distance to mate score. So all it sees a tablebase loss, and one losing move is as good (or as bad) as another.
