No one else tried to implement Leela themes?
Constriction: What I've learned from Leela and Alpha Zero
Sort:
I think it's interesting the way Leela and Alpha Zero play, however, I'm not going to copy the way a 3400 rated engine plays when I'm competing against 1600 rated humans.
I'd much rather copy the way a FM/IM/GM plays, because humans tend to make more natural moves that are easier to follow and easier to understand intuitively.
Deranged wrote:
I think it's interesting the way Leela and Alpha Zero play, however, I'm not going to copy the way a 3400 rated engine plays when I'm competing against 1600 rated humans.
I'd much rather copy the way a FM/IM/GM plays, because humans tend to make more natural moves that are easier to follow and easier to understand intuitively.
I think Leela plays rather human though, which is why I'm studying it's playstyle.
Given that Neural Network or Monte Carlo AI has already helped you in your own play, I'd say, "Carry on!"
Feb 16, 2019
0
#6
I didn't see the first game, but I looked at the second - your opponent blundered a pawn, and you took it - a center pawn. After you're a pawn up, you are just winning, and there is no reason to complicate the position with Rxg6, which is a blunder, because hxg6 Rxg6 Kf7 is winning for black - white has only two pieces in the attack, because the Ne4 blocks the queen and bishop, and is safely defended. The appropriate thing to do is to just retreat your rook, then blast open with h5 - no sacrifice needed.
In fact, after the sacrifice + ...Kf7, your opponent will quickly attack you, by playing ...Rg8, etc.
Computers need to sacrifice material against other computers because computers are extremely good at not just losing for free, whereas club players are hanging stuff left and right.
The fact that you played a bad move that took you from winning to losing, and your opponent played worse ones, does not mean the bad move was sound.
It's like me beating up 1500s by playing horribly but flashy, and thinking I'm a GM or something.
I think the idea of emulating engines is insanity, and oddly enough, only comes from players who have never even seen the source code of an engine, or implemented an engine themselves. Its not magic - it's using mass computational power to brute force a game with finite possibilities.
No offense, but I've seen this kind of thing a lot. It never works, and is usually a huge waste of time.
RussianHAMMER wrote:
I didn't see the first game, but I looked at the second - your opponent blundered a pawn, and you took it - a center pawn. After you're a pawn up, you are just winning, and there is no reason to complicate the position with Rxg6, which is a blunder, because hxg6 Rxg6 Kf7 is winning for black - white has only two pieces in the attack, because the Ne4 blocks the queen and bishop, and is safely defended. The appropriate thing to do is to just retreat your rook, then blast open with h5 - no sacrifice needed.
In fact, after the sacrifice + ...Kf7, your opponent will quickly attack you, by playing ...Rg8, etc.
Computers need to sacrifice material against other computers because computers are extremely good at not just losing for free, whereas club players are hanging stuff left and right.
The fact that you played a bad move that took you from winning to losing, and your opponent played worse ones, does not mean the bad move was sound.
It's like me beating up 1500s by playing horribly but flashy, and thinking I'm a GM or something.
I think the idea of emulating engines is insanity, and oddly enough, only comes from players who have never even seen the source code of an engine, or implemented an engine themselves. Its not magic - it's using mass computational power to brute force a game with finite possibilities.
No offense, but I've seen this kind of thing a lot. It never works, and is usually a huge waste of time.
I already stated in the OP that the rook sac registers as a blunder under analysis. It's a 3 minute blitz game and blacks defense is not readily apparent at our rating and time control.
What I think is insane is that you critique a game on perfect engine analysis post-mortem.
Feb 17, 2019
0
#8
Black's "defense" is an attack - ...Kf7 immediately brings the Rg8-g1+ threat, winning the queen. When you're up a center pawn, winning, you don't make wild sacrifices that might or might not work - to praise that is like praising a kid for doing all the wrong calculations on a math problem, and getting lucky on the right answer. Going for Qh5+Bc4 against an 800 rated might work - doesn't mean it's good. Playing Qh5-Qxf7+ and hoping your 400-rated opponent will just resign instead of taking the queen might work - still not good.
This is far from perfect engine analysis - I didn't actually use an engine to analyze ...Kf7. You don't know how terrifying engines are if you think this is anything, though.
When I was like a 1200 OTB rated player, I thought my flashy games were somehow masterpieces, until I realized that they would've cost me the game against anyone higher rated. I'm only saying this because I actually can relate - obsessing with engines is a complete of time.
If I say "nice win!" and you ate it up, you'd be a worse chess player because of it. The fact is Leela has nothing to do with anything you played. Leela calculates hundreds of millions (or billions?) of variations before making a move.
If you don't believe me, well, whatever.
I've been watching Kingscrusher videos on Leela (and other people on Alpha). What I've gathered from watching them is that Leela aims to constrict enemy movement, in particular the queen's side bishop and the queen side rook.
Once Leela has established the the opponent is virtually down one or two pieces, due to constriction, and has pointed all of its pieces at the enemy king either directly or via x-ray/discovery, it has no problems sacrificing pieces and pawns in order to "blow up" the position and either checkmate the enemy king, or to force the opponent to sacrifice so much more material in return to prevent the mate.
I wish to highlight how these "constrain, ATTACK!" playstyles of Leela have inspired my gameplay in two of my recent games:
https://www.chess.com/live/game/3462635094
In the above gave, move 12 Nf3 and then 13 Nd4 aim to blockade black's d pawn and "constrain" his queen side bishop. On move 15, I've determined it safe enough to push my king side pawns and I am considering moving my king to the h-file to double my rooks on the gfile (something that leela and alpha both do). 17 Bg3 allows me to move the f pawn so I can rook lift on the f-file or double rooks on the f-file. On move 20f5 My white bishop has extraordinary mobility and black's white bishop remains constrained, therefore f5 allows me to initiate a strong white square attack. On 24.Kg2, I know that my opponent's white bishop cannot threaten my king with the x-ray because of how secure my blockade is on the d file, I prefer instead to ensure that he can't x-ray my king with his queen on the dark squares and his rook has an x-ray on the h file, 24. Kg2 seems to be the best move for the long term. Black blunders on move 30, hanging his queen.
https://www.chess.com/live/game/3459513123
In my victories between this game and the first game I posted, my opponents blundered pieces, so they aren't worth looking at. Now for this game:
On move 23: Rhg2 I am prepared to make the exchange of my first rook for his knight, black has four pieces that cannot defend his king (in time) against my h-pawn, 2 rooks, black bishop and whitebishop+queen battery (x-raying the knight he intends to capture my rook with).
On 24. Rxg6 (epic rook sac) I am confident that I have more than enough pieces pointing at his king to justify the rook sac, since I only need to push my f-pawn to force his knight to move and open up the queen-bishop battery on the white squares. Stockfish shows 24. Rxg6 to be a "mistake," yet, black's defense against this attack is actually quite complicated to pull off in order to capitalize on this "mistake," in a 3 minute blitz game. That being said, my "leela gut" told me to go for this rook sac.
Black soon falls apart to this devastating attack in only a few moves since his pieces were too far away from the action (constraint).
Also just wanted to say thank you to chess.com for banning the sandbaggers and cheaters at the entry level bracket and refunding our points. I'm guessing you straight up IP/hardware banned them this time because I haven't seen any since that ban wave. Seriously thank you, it was making the supposed "entry level" bracket unplayable.
Thank you!
Enjoy
Link to Kingscrusher channel:
https://www.youtube.com/channel/UCDUDDmslypVXYoUsZafHSUQ
These games are really worth studying in my opinion, they really demonstrate human-like chess, but done "correctly." Watching regular computers like Stockfish doesn't really teach you anything except that human begins suck at short-term tactics lol.