< Earlier Kibitzing · PAGE 15 OF 15 ·
|Jan-06-19|| ||AylerKupp: <<keypusher> Re one of your concerns about AlphaZero -- that it seems to reach 3500 Elo rather quickly, then plateaus -- I wonder if it's bumping up against a draw ceiling in chess.>|
Perhaps. It's hard to tell at the scale of the charts what AlphaZero's Elo rating gain was when going from about 150K training steps (when it first reached the approximate 3400 Elo rating) to 700K steps. But I can say that in the latest (Jan-04-19) CEGT 40/4 tournament (6 secs/move) Stockfish 10 4CPU gained 138 Elo rating points (about 4%) over Stockfish 8 4CPU when running on comparable hardware. So it seems that Stockfish is still capable of improving, at least at fast time controls, although I don't know how long it can keep this up.
I don't know if Elo ratings are not comparable across games since they are based on relative performances, and of course you can't compare AlphaZero's performance against Stockfish with AlphaZero's performance against Elmo or AlphaZero's performance against AlphaGoZero. And I don't know enough about either Shogi or Go to even make a guess.
|Jan-06-19|| ||ChessHigherCat: <keypusher: Re one of your concerns about AlphaZero -- that it seems to reach 3500 Elo rather quickly, then plateaus -- I wonder if it's bumping up against a draw ceiling in chess. One thing that stands out in the Science article is how common draws are in chess compared to shogi or Go.> |
It's probably premature to talk about the death of chess but I bet what will happen is that the computers will refute (or "bust") certain dubious openings so that nobody will play them anymore. And maybe certain openings will be proven to be inevitable draws given optimal play on both sides, so white would stop playing them too (except against a much stronger player to avoid getting scrunched).
|Jan-06-19|| ||john barleycorn: <keypusher: Re one of your concerns about AlphaZero -- that it seems to reach 3500 Elo rather quickly, then plateaus -- I wonder if it's bumping up against a draw ceiling in chess. One thing that stands out in the Science article is how common draws are in chess compared to shogi or Go.> |
Because "draws" in chess can be "agreed" (almost) any time and not solely be so by the position on the board but by other considerations. Then hitting a draw ceiling would mean the strongest engine would lose rating points, won't it?
|Jan-06-19|| ||keypusher: <<CHC> it's probably premature to talk about the death of chess but I bet what will happen is that the computers will refute (or "bust") certain dubious openings so that nobody will play them anymore. >|
Looking at recent top level events it feels like thatís already happening. Some of us are hoping A0 will counteract that, at least temporarily. If it can sacrifice three or four pawns against Stockfish and get away with it, maybe thereís more life in the game than top GMs seem to realize.
|Jan-06-19|| ||SugarDom: Is King's Gambit a "bust" for White then?|
|Jan-06-19|| ||WannaBe: King's Gambit is a bust, but Queen's gambit are busts!|
Ok, I'll move along now....
|Jan-06-19|| ||ChessHigherCat: <WannaBe: King's Gambit is a bust, but Queen's gambit are busts!>|
Can't we keep abreast of recent developments without such crude puns?
|Jan-09-19|| ||offramp: Has anyone seen any really outlandish openings in recent computer games, that are at chessgames.com?|
I have seen an Evans Gambit, and a Two Knights (the one where White plays Ng5 and Black plays ...b5), but those are quite standard openings...
I am thinking more of games like Trompowskys, Muzios, Cochrane Petroffs, Goring Gambits, Frankenstein-Draculas. These openings are theoretically reasonable, so they might have been included in the ballot.
|Jan-09-19|| ||keypusher: <Offramp>
Here's a Frankenstein-Dracula (apparently called the Blanel Gambit in more refined circles).
AlphaZero vs Stockfish, 2018 (Draw, 135 moves)
Stockfish vs AlphaZero, 2018 (Draw, 194 moves)
King's Bishop's Gambit:
Stockfish vs AlphaZero, 2018 (Drawn in a brisk 56 moves)
Finally, a Fried Liver:
Stockfish vs AlphaZero, 2018 (Draw, 198 moves)
Only the 10 games released in 2017 seem to be accessible via opening explorer. I'll register a complaint with the proper authorities.
|Jan-09-19|| ||Diademas: The more It watch Alpha0's games, the more I want to give up chess.|
|Jan-09-19|| ||john barleycorn: < Diademas: The more It watch Alpha0's games, the more I want to give up chess.>
Why? you are not forced to play the devil|
|Jan-09-19|| ||Diademas: <john barleycorn: < Diademas: The more It watch Alpha0's games, the more I want to give up chess.> Why? you are not forced to play the devil>|
I'm afraid I struck a deal with the devil in the early 80's that I can't withdraw from...
|Jan-09-19|| ||john barleycorn: < Diademas: ...
I'm afraid I struck a deal with the devil in the early 80's that I can't withdraw from...>
Well,then good luck.
|Jan-09-19|| ||Diademas: Thanks <John>, I might need it.|
|Jan-09-19|| ||offramp: <Keypusher>, many thanks. That is exactly what I was looking for. Those games are very interesting!|
|Jan-10-19|| ||offramp: Another Fried Liver attack, a win this time.
AlphaZero vs Stockfish, 2018, 55 moves.
|Jan-10-19|| ||Tiggler: <Unless you run 2 software programs on computers of similar performance you can't reach a definite conclusion as to which software programs is the superior performer.>|
But if one engine is unable to use state-of-the-art hardware, that is hardly a reason to be skeptical of the one which can.
|Jan-10-19|| ||keypusher: <offramp> Thanks, cool game!|
|Jan-16-19|| ||scholes: Alphazero claimed to beat tcec champion stockfish. |
Alphazero little open source sister leela aka lco breaksthrough at TCEC. Qualifies for TCEC superfinal against stockfish ahead of Komodo and Houdini 6, by winning division 3, division 2 and finishing 2nd to stockfish in div premier.
It is also currently at 2nd place in cccc
|Jan-17-19|| ||zanzibar: I'm going to release a rationalized version of the recent AlphaZero--Stockfish games, first focusing on the fixed opening games.|
If all the games were available, I'd divide the match into a series of matches, one for each opening (which I think was 20 games/color(?)). In other words, each game for an opening would have its own Event tag. That would make getting the actual stats easy.
(Aside - does anybody know a table of the numbers, as compared to the colored bars in the supplemental. I'd like to know the numbers, instead of eyeballing 'em).
I'm also thinking of removing the misleading "book" comments, and instead use the FEN tag for the PGN, putting the opening moves into a comment.
That would have the advantage of clarity (ie. what moves the engines actually made). It would also eliminate having to play over the same opening moves in the 40 (?) match games.
Hopefully this time round we'll be able to get all the games (now that DeepMind has gotten its paper published).
Anyone else have thoughts on this, particularly on using FEN to specific the starting position?
|Jan-17-19|| ||zanzibar: Does anybody have a computer-readable file of the TCEC-2016 openings?|
A PGN or even a list of FEN's would be nice.
And can someone tell me when the openings were used? Was it in the superfinal stage only?
(Yes, I might be able to dig that out, but if someone could provide a quick pointer I'd be obliged.)
|Jan-18-19|| ||zanzibar: Best way to sort the TCEC opening games?
1) ECO code?
2) By S4 diagram?
|Jan-18-19|| ||zanzibar: RE: <Does anybody have a computer-readable file of the TCEC-2016 openings?>|
Has < [ 2006, 2008, 2012, 2012 (Topical), 2014, 2016 ]> Noomen Testsuites.
(Following the "canonical" link from chessprogramming/etc yields a stale mediafire link:
I like this one opening example PGN, which provides a fuller description than Noomen's.
I'm trying to fill out the PGN info similarly.
|Jan-19-19|| ||keypusher: <zanzibar> I have nothing to offer you but best wishes -- I suspect you can do more for yourself than most of us can do for you. I've ordered <Game Changer>, and I'll pass on anything l learn. But I don't think I'll get it for another month.|
|Jan-19-19|| ||zanzibar: Thanks for the good wishes, <kp>. |
I miss the days when <Pawn and Two> and <PaintMyDragon> (and others, of course) offered valuable help. But I guess I've brought it upon myself, trying to tame the PGN wilds.
< Earlier Kibitzing · PAGE 15 OF 15 ·