lkaufman wrote: ↑Wed Oct 16, 2019 6:43 am
I think I'll change my opinion from "draw" to "loss" for 1.g4. I let Komodo MCTS think about it for an hour or so and got a -.83 score for it after 1...d5. Usually anything beyond 0.7 is a winning advantage based on playouts from the position. This doesn't mean that I or anyone else can prove the win, just that it is probably the smart way to bet if somehow there were a way to get the answer!
What do you consider proof?
Regards,
Zenmastur
a tree that show a forced mate in all lines is certainly a proof.
1.f3(f4) e6(e5) 2.g4 is proved to be a forced mate for black(white can use a different order of moves)
1.f3(f4) e6(e5) 2.h3 is also proved to be a forced mate for black(white can use a different order of moves)
I am not sure if you can find a proof for a forced mate after 3 plies except these examples
There are some lines when white lose the queen like
1.e3 Nf6 2.Qg4 or 1.e4 g6 2.Qh5
I am sure that black win in them but I am not sure if you can prove a forced mate for black in one of them
Well...
I know what a proof tree looks like. I also believe that with current hardware generating a proof tree for 1.g4 is out of the question.
I'll be happy with any score of -3.xx or less (since that's what I recalled/claimed).
Regards,
Zenmastur
Only 2 defining forces have ever offered to die for you.....Jesus Christ and the American Soldier. One died for your soul, the other for your freedom.
Ovyron wrote: ↑Mon Oct 14, 2019 5:38 pm
I'm willing to defend 1.g4 against anybody that seriously think it is losing, and I guarantee that I can draw them.
Am willing to give it a go 1...d5. if you want to play we should probably play the game in the tournaments and matches sub-forum?
Ovyron wrote: ↑Mon Oct 14, 2019 5:38 pm
I'm willing to defend 1.g4 against anybody that seriously think it is losing, and I guarantee that I can draw them.
Am willing to give it a go 1...d5. if you want to play we should probably play the game in the tournaments and matches sub-forum?
Ovyron wrote: ↑Mon Oct 14, 2019 5:38 pm
I'm willing to defend 1.g4 against anybody that seriously think it is losing, and I guarantee that I can draw them.
Am willing to give it a go 1...d5. if you want to play we should probably play the game in the tournaments and matches sub-forum?
Ovyron wrote: ↑Mon Jan 13, 2020 5:15 am
But what do you think Harvey? Is g4 defensible? If so, at what point did I go from drawing to losing?
I think g4 is probably losing. I think possibly another move 2 could be worth a try. Uri posted analysis somewhere that according to Stockfish 2. c4!? Is worth a look. I can’t find any games where this has been tried.
Ovyron wrote: ↑Mon Jan 13, 2020 5:15 am
But what do you think Harvey? Is g4 defensible? If so, at what point did I go from drawing to losing?
I think g4 is probably losing. I think possibly another move 2 could be worth a try. Uri posted analysis somewhere that according to Stockfish 2. c4!? Is worth a look. I can’t find any games where this has been tried.
I think the short castling was the big error of white. I base that on my very very bad chess skills
A while back I was experimenting with a book learning system, and for a while it was selecting a number of "bad" moves including 1. g4. What I found was that even with poor opening choices the draw rate with Stockfish (self-play) was very high. This kind of screwed up the learning because poor choices weren't "punished" with bad outcomes. Playing against another engine would probably reduce the draw rate, and in a match-up with a stronger engine playing 1. g4 might be a bad idea. But it seems defensible with a strong engine. 1. b4 was ok, too.
jdart wrote: ↑Mon Jan 13, 2020 5:08 pm
A while back I was experimenting with a book learning system, and for a while it was selecting a number of "bad" moves including 1. g4. What I found was that even with poor opening choices the draw rate with Stockfish (self-play) was very high. This kind of screwed up the learning because poor choices weren't "punished" with bad outcomes. Playing against another engine would probably reduce the draw rate, and in a match-up with a stronger engine playing 1. g4 might be a bad idea. But it seems defensible with a strong engine. 1. b4 was ok, too.
--Jon
Sounds like a weak move selection algorithm, or not using all the knowledge you have available, or too high expectation of the learning rate, or a combination of all three.
I'm kind of curious what was the time controls you were using? I think this plays a huge role in learning rate per unit time spent.
Regards,
Zenmastur
Only 2 defining forces have ever offered to die for you.....Jesus Christ and the American Soldier. One died for your soul, the other for your freedom.
Sounds like a weak move selection algorithm, or not using all the knowledge you have available, or too high expectation of the learning rate, or a combination of all three.
I'm kind of curious what was the time controls you were using? I think this plays a huge role in learning rate per unit time spent.
I didn't get very far with this but I was using a fairly long time control. Shorter TC would give less of a draw rate and faster learning but also I think more randomness in the results. The problem is, I think even in the best case, you need a very large number of games to get convergence. In a lot of cases one move scores 60% and another 55%. That is significant but getting to the point where you see that difference is going to take time. You could also optimize for scores out of the search, which I think a lot of people have done, but IMO that is less reliable. For example, something like the Ruy Lopez Marshall Gambit where Black is a pawn down may give you minus scores, but most of the endgames are drawn.
lkaufman wrote: ↑Mon Oct 14, 2019 9:41 pm
I think that the best we can do at the present time to tell if a given opening position (not one that is obviously apt to be solved by brute force) is a win or a draw is to run the strongest version of Lc0 on it for a few minutes on a 2080 or better GPU with win prob. display. The threshold for a won position is probably about 70% or perhaps slightly higher, 71 or 72%, based on everything I've observed. For 1 g4?, it is about 68% for Black with 1...d5, so probably not a forced win although close enough to the line to leave that as a possibility.
Leelenstein 13 already is below 30% winrate after g4 d5. By 10 million nodes it has switched from preferring g2 to c4 (which someone else mentioned as possibility), when it saw how bad g2 was, and then c4 dropped some more too. With more time its probably easy to get to your 71% threshold.