The Leela Experience

My theory is, that playing against Leela with piece odds will help me convert winning positions I get OTB. I'll be using this thread to post my thoughts about the games I'll play, games itself and really anything that comes to mind.

First I played 2 games against @LeelaKnightOdds, that didn't go well, so I had to downgrade my opponent to @LeelaQueenOdds. How hard could it be? I lost the first 7 games.

Leela is really good at starting an attack before I castle and at giving pieces a very long term task.

The turning point came when I tried out playing a gambit

Therefore I'll try to focus on playing normally, as that's what I clearly struggle with still.

Another thing is, I tried to adapt Leela's style of very slowly building attacks and putting pieces on squares, where trading them further benefits my position.

With 10 losses on my back I think it's time for some analysis. First I want to talk about stuff I did right when I won, then compare that to games where I lost and also focus on what does Leela do to win so consistently. I will not focus on the openings almost at all, unless there is an idea I want to talk about.

Game 1

As I said before, I won 4 games with gambits, while playing normally I'm ½-10½. I won't go over full games, instead focusing on the key ideas and how they repeat along different games.

Piece play

Focusing on pieces rather than material is what all these games have in common. In the first game I was able to pull off a textbook simplification

Simplifying 1 / Calling the bluff

Just because Nd5 looks scary doesn't mean it is.

Simplifying 2

Simplifying 3

It's kinda rare to get so many trades against Leela, so focusing only on this won't get me far. You can see how she refuses to take a free knight on c2 just to keep a rook on the board.

Game 2

Similar start as the last game. This one shows it's important to not take Leela's moves as granted and calculate and evaluate them yourself.

Calling the bluff

Game 3

Piece play 1

Even against f4, the same concept. The point is not to play the From's Gambit and go for White king, but focusing on the pieces instead.

Piece play 2

Doesn't matter what phase of the game it is, prioritizing development is key.

Attack

Piece development opens the doors for stuff like this. Even with no clear follow up if we have enough attackers something will come up.

Game 4

With the same 1.d4 e5 opening as before we can launch attacks pretty early into the game

Attack

And the king's lack of safety is pretty important.

Simplifying

All thanks to Qg6+.

It seems there is a limit to how many boards can be displayed in a single forum message, which is 10, but there is one more game, where I did a lot of stuff right, which I'll show separately.

EDIT: The limit is about links. You can post 10 links per day in the forum, therefore only 10 boards in one message

Game 5

If you've played Leela you know how she generates these slow attacks out of thin air. There are 2 ways of dealing with this: looking for faster counterplay on the other wing, or in the cenrte; or defending actively.

Defensive outposts

It becomes much harder for White to attack and we can use that time to generate play somewhere else.

Pawn breaks

With the classic to finish off the game

Simplifying / Calling the bluff

It's time to talk about the hard part — the losses. For now I have 10 under my belt and I think it's a really good moment for conclusions. 15 games should be enough to see what I did right and wrong and I hope after the next 10 games the score will be much more in my favor.

After going over the 4 wins and an unfortunate draw I can already see the improvement. When analyzing these defeats, I see what I'd play differently, even in positions where the engine isn't really helpful.

Let's start from the first loss.

Game 1

I think the biggest takeaway is piece play. On move 13 I have 3 pieces developed and they stand on awkward squares: knight on d7; bishop on g7 looking at a pawn chain and the queen on a5 with no targets. Also castling into two bishop can't be a good idea.

I'm not sure why I resigned so early.

Game 2

I think playing d5-c5 is much easier to handle against b3, but still Black has a great position, the problems start after 6...h5!? when I shift my focus to pawn grabbing rather than development.

It becomes clear after 21.Nxe4 that I did something wrong.

Game 3

Closing the position with 8...d4 is controversial. The Benoni structures tend to be much complicated and the e4 pawn break isn't easy to achieve.

I lost a pawn early on, but had I played for f5 the position would be amazing. Of course, the a6-b5 plan isn't bad by any means, f5 was just better.

White got a strong pawn on c5 but I was able to defend correctly, win it and simplify into a winning endgame, sadly losing on time.

Game 4

While 2...c5–e6 aren't in the spirit of piece play, I handled the opening very well with a quick b5-c4. However I didn't use the tension to my advantage very well, missing chances to exchange the pawns. Luckily for me a queen up is still a queen up and even in a closed position I was able to break-through. Sadly, the clock wasn't my friend again.

Game 5

I don't know why I closed the position again, but at least I handled it well again opening the kingside with 15...g5! and finding great squares for my pieces, which makes it crazy I lost such an active position.

From this game I think we can learn that patience is key. There was no ruch to play 30...Rxc7, I should've started with Ne5 instead.

Game 6

A perfect example of what not to do. After finishing development I literally did nothing and Leela could play freely on the kingside. Also I resigned early for some reason.

Game 7

Again, I'm not a fan of 2...e6 and 5...d4. Kinda sad I didn't find the perfect square for the f8-bishop with 13...g6–Bh6, but I used the bishop on d6 quite well, opening it up and parking it on f4.

That sums up all the things I did right. Blundering a piece like that is ridiculous, considering even after missing 20.e5 I still could give up the queen for two pieces.

Game 8

Solid opening made it hard for Leela to find anything. 22.Nc5 allows me to simplify with 22...Bxc5, but even after inaccurate 22...Qc8? blundering a piece, I still managed to realize that a sacrifice on e6 is necessary.

In the resulting endgame all I have to do is activate the knight, but after one inaccuracy I wasn't able to find a plan and ultimately lost the game.

Game 9

I wanted to try something else... In the final position I have 3 pieces on their home squares, that might've been a problem.

Game 10

Similar story. I tried to go for an attack with no development and got punished for that.

Conclusion

I think the two most important lessons are focusing on piece play over anything and always being ready to call the bluff and simplify.

I also have some good news. After the analysis I played the bot again, winning 4-0 with 10+5 time control, then another 4 in a row playing 5+3 and finally in 3+2 I won one and lost one. The rapid games were too easy, so I'll be moving on to blitz and soon to smaller odds.

In 5+3 games I've missed some stuff, but had enough time to figure things out. I think I'll still mix both time controls until 5+3 feels too easy aswell. I'll post the analysis soon.

GOOD NEWS! It's time to move on to the harder bot @LeelaQueenForKnight as with queen odds after the analysis of my first games I went 17-4, most importantly losing only with 3+2 time control, but even there the score is 8-4. I don't think there is much to gain from analyzing full queen odds past this post. From now on I'll be playing @LeelaQueenOdds either for fun or as a warm up, or maybe try to beat it in 1+0 bullet.

Let's go over the first 10 games after I did the analysis and see the improvement!

10+5 #1

Insanely patient play from me simply preparing c4-break from 12th to 24th move! And after the position opened a nice maneuver Nb7–Nb4 to open the c-file. Was it this simple all along?

10+5 #2

I thought it was interesting to close the position on the 3rd move, but my point was White had to chose between locking the f1-bishop or misplacing the e2-knight. When I was able to recapture on e5 White had no play. I'm surprised Leela didn't opt for f5, or g3 instead.

Again, very patient play after Leela created a hook with 18.h3, simply preparing the pawn-break.

10+5 #3

It seems like putting the bishop on g7 isn't a good idea against the engine, as the games always seem less clean. Missing 19.Ng1! didn't help as well, but I'm happy with my play after blundering a piece.

10+5 #4

I felt like I got my pieces tangled this game quite bad, but somehow it all worked out when the position opened up. After this game I decided 10+5 is too easy, so I switched to 5+3.

5+3 #1

Probably the cleanest game up to this point: quick development; stopping Leela's e2-e4 and a deadly attack to finish it off.

For whatever reason Leela decided to repeat the game again

5+3 #2

With even faster mate

5+3 #3

Similar story, quick development and a sacrifice, not much to say.

5+3 #4

Opposite side castling is always interesting against Leela, as I have to show some defensive skills. In this game putting the knight on h5 and later on e6 was enough, as my counterplay on c2 was MUCH faster and dangerous. After this win I decided to switch to 3+2, as this tempo was also too easy.

3+2 #1

Similar to some 10+5 game before, I was just preparing the c4-break. All I had to do on the kingside is meet 21.g4 with g6 and I could play freely on the queenside.

3+2 #2

I think the only game where I just won a piece somehow I managed to throw. First of all I had mentally given up after 19.Be2. Instead of looking for the best way to give up the queen, I just captured the bishop. The same was true after I got my bishop trapped with 26.c5.

Another thing is I lacked patience in the endgame. It's clear what was White's play and that it's faster than mine, so I should first prevent that, only then focus on my asset. More specifically starting with f5 and Nd7 instead of pushing the h-pawn.

Lesser odds

Obviously I expected this to be harder, but damn. Remember, I can beat QueenOdds 2-1 on average in 3+2 with Black. I played QueenForKnight twice, in 10+5 with White and lost twice. Even though I was expecting a challenge, I'm still surprised. The way it can outplay me with a rook instead of a queen is crazy. I wonder how long it will take until I'll get half decent against it.

game 5 where it has the sentence "I don't know why I closed the position again, but at least I handled it well again opening the kingside with 15...g5! and finding great squares for my pieces, which makes it crazy I lost such an active position." (ctrl+f) there was 28... f5

@StevenEmily said ^

game 5 where it has the sentence "I don't know why I closed the position again, but at least I handled it well again opening the kingside with 15...g5! and finding great squares for my pieces, which makes it crazy I lost such an active position." (ctrl+f) there was 28... f5

Yeah, that would make things much easier, nice spot!