Amazing results of LeelaKnightOdds-Dev

[lkaufman]

I find it difficult to explain to my non-chess friends what it feels like to see these queen odds games.
Finally I came up with a way to express this (subjective) feeling: someone came up with a way to win at tic tac toe. What??? Exactly.

Cheers,
Peter

I like that, although it's not quite comparable because Leela "only" found a way to beat maybe 99% of the world's chess players giving them queen odds, not an actual win. So imagine someone coming up with a way to beat 99% of humans in tic tac toe, but not "grandmasters" of tic tac toe!

[Uri Blass]

Interesting what is the change of other players against the bot.

BvAR2 played in 1.11-11.11 before the upgrade 195 games with white against the bot at 15+10
The results were 111 wins for leela and 75 wins for BvAR2 that means that he got 79.5 points out of 195(40.77%)

Same player played 79 games against the upgrade and won 15 and lost 58(6 draws)
meaning 18 out of 79(22.78%).

It seems that when the difference against me is more than 200 elo the difference against BvAR2 is clearly less than 200 elo.

[Uri Blass]

Note that I can probably do better against leela queen odds by some good book to memorize and I already started to build a book based on the last game in order not to repeat my mistakes when I consider every move that I played and Wasp450 does not suggest as a mistake(my book include 39 moves).
I already played the last games based on the book and of course I need to repeat the moves of the book to memorize them but I can memorize some hundrends of moves in a repertoire by training if I want to do it.

Maybe it is better to try to memorize the moves I played in games that I won and repeat them and only change the moves in games that I did not win in some point of the game but I got positions that I did not like even in games that I won and I think there should be a simpler way to win.


For example my last game
https://lichess.org/TJyAX0IV2lVj

In this game I sacrificed my queen by 11...Nc6 in order to trade a lot of pieces.
I did not feel the way to win is simple after sacrificing the queen and I did not like it but decided that I have no better option because otherwise white has the initiative.

I was leading by material 38-28 and after lc0 accepted the sacrifice I was leading 23-17 so it did not improve my ratio.
38/28=1.357 when 23/17=1.3529...(of course did not calculate it during the game).

Later I again sacrificed by choosing to lose the exchage when I did not like it by 19...Rd5 because I felt that I have no clear plan how to win with my pawn structure.

Not sure if it is a practical correct decision of lc0 to accept the sacrifice and analyzing the game I see that I do not lose a pawn if I choose not to lose the exchage because I have ideas of Rd2 against Na5 if I play simply Kf7 but in the game I was afraid of taking c6 with a fork.

[Father]

I find it difficult to explain to my non-chess friends what it feels like to see these queen odds games.
Finally I came up with a way to express this (subjective) feeling: someone came up with a way to win at tic tac toe. What??? Exactly.

Cheers,
Peter

Good morning Peter. I feel transported to London and Berlin in the mid-19th century when I watch every game LeelaQueenOdds plays; the computer seems to draw a work of art with a brush, emulating the lines of play of the greats, Adolf Anderssen (white) and Lionel Kieseritzky (black) in London on June 21, 1851, and Adolf Anderssen - Jean Dufresne, 1852 in their game played in Berlin in 1852. LeelaQueenOdds could very well have the aliases of "The Immortal Machine" or "The Everlasting Machine". I too cannot find a way to explain to myself this new era of Odds chess. On the other hand, I am noticing that titled players are coming to the ring more and more. All this makes the battle between four worlds great and relevant. Happy day Peter.

[Uri Blass]

For some reason I cannot find Punypawn315 when I search for this player in lichess.
This player played only 19 games but already has 1952 that is very good considering the fact that he is supposed to start with 1600
8 Punypawn315 1952.0 19

[Uri Blass]

I can add that there is a problem and for some reason leela does not play lately.

She accept games but do not make a move.
There are a lot of games that players aborted and I can see that in theory she is playing 10 games now but practically did not play the first move.

https://lichess.org/@/LeelaQueenOdds/playing

[Uri Blass]

Now I see that leela won some game on time after 1 move and I hope you do not count them for the rating list.

[Marcus91]

For some reason I cannot find Punypawn315 when I search for this player in lichess.
This player played only 19 games but already has 1952 that is very good considering the fact that he is supposed to start with 1600
8 Punypawn315 1952.0 19

He scored 55% at 1+1 just before the upgrade, in the first 20 games 'K' is set to 40, so grow is quite fast, everything is in order.

[Marcus91]

Now I see that leela won some game on time after 1 move and I hope you do not count them for the rating list.

Ok, I will add a check that a game must last at least 5 moves

[Uri Blass]

won another game at 10+5.
Third win in a row against her.

I needed to sacrifice material and not sure if she did correctly when she captured it or not.
It was not an easy win in time trouble even after she captured the material but maybe better chances is not to capture the material.


I wonder if it learn from games against opponents about specific weakness of the specific opponent and I think that it can be also a possible improvement.

I think that I should go back to second place after this game is going to be included because I had 2119.6 in the rating list before the game when father had only 2123.