I haven't been in the community for a while. Has any recent progress been made in solving atomic?
Do you know of someone who'd be working on it ?
What even do you believe to be the result, win or drawable ?
Problem is, if it's drawable, it might never get solved ^^ (I mean not before chess is)
What even do you believe to be the result, win or drawable ?
Problem is, if it's drawable, it might never get solved ^^ (I mean not before chess is)
Atomic is imbalanced and Stockfish gives +2 for the starting position,so white is better.Is it a win for white or not,i don't know,but atomic player @savagechess2k tried it by making stockfish play itself and white won.
@lecw I'm probably going to be able to access supercomputers and people with similar interests this fall. That means if possible I will start the project.
#4 good luck bro
@kasparov_4 Thanks! I'm learning about deep learning and am likely to be able to work in data science starting from this fall. I'm currently looking for others who are interested in the project.
Where are you going in the fall ? I did half my masters at a deep learning lab.
@cat_person Post the updates on this forum please: lichess.org/forum/team-atomic-chess-theoreticians In the general forum this would be hard to find for people who are interested.
I'm quite interested in Atomic solution, but I think there is no certainty in the community it is a win. Black has some tenacious defences in pretty any line white can play, and everything is complicated by the fact there are several miraculous Atomic drawish patterns that Stockfish doesn't understand and evaluates as +3-+4 which is higher than the eval of initial position.
If you're interested in deep learning, probably it would be nice to create Leela Atomic Zero first which will probably understand such things as pawnitization or dead drawn exchange-up endgames better, and use it to build the perspective solving tree from the opening. However, building of such full tree for the whole game as was done for Antichess is doubtful to be computationally feasible, as Atomic is considerably less forcing game.
As for the concrete progress for now, I saw good attempts in 1. e3 e6 2. Nf3 line in @Xeransis vs @AtomicChessBot games in the Lichess database. This is the only line that feels relatively close to be solved informally, but it is still far from the formal proof. I should probably warn you to not waste your time for the most popular "winning" line 1. Nf3 f6 2. e3 d5 3. Ng5 unless you have an engine with better understanding of the draw concepts mentioned above. Atomic is FAR from being solved by this line, although it's pretty unpleasant to play against.
I'm quite interested in Atomic solution, but I think there is no certainty in the community it is a win. Black has some tenacious defences in pretty any line white can play, and everything is complicated by the fact there are several miraculous Atomic drawish patterns that Stockfish doesn't understand and evaluates as +3-+4 which is higher than the eval of initial position.
If you're interested in deep learning, probably it would be nice to create Leela Atomic Zero first which will probably understand such things as pawnitization or dead drawn exchange-up endgames better, and use it to build the perspective solving tree from the opening. However, building of such full tree for the whole game as was done for Antichess is doubtful to be computationally feasible, as Atomic is considerably less forcing game.
As for the concrete progress for now, I saw good attempts in 1. e3 e6 2. Nf3 line in @Xeransis vs @AtomicChessBot games in the Lichess database. This is the only line that feels relatively close to be solved informally, but it is still far from the formal proof. I should probably warn you to not waste your time for the most popular "winning" line 1. Nf3 f6 2. e3 d5 3. Ng5 unless you have an engine with better understanding of the draw concepts mentioned above. Atomic is FAR from being solved by this line, although it's pretty unpleasant to play against.
@lecw Seattle. If you are in the region maybe we can even have a meet up after May.
@Wolfram_EP Sure! I'm in fact a fairly old member of the team. However this time I may actually be able to make concrete steps towards the solution of atomic.
Yep I'm already talking to Leela people. This will be a cool deep learning project.
Yep I'm already talking to Leela people. This will be a cool deep learning project.
This topic has been archived and can no longer be replied to.