Opening Choice by Leela

For those who are interested to know the opening choice by current world best chess player Leela Zero( Both Stockfish and Leela are best two programs with equal strength in comparable hardwares but  Leela is better in opening and SF is better in ending).

T60, best choice in order of popularity,

1. d4

2. Nf3

3. e4

Thanks

If you have extra moments, how does Leela rate the McCutcheon French or French in general?

French is too general. Just give me 4 or 5 variations of French or FENs.

I will analyse mainline of French MuCutcheon., this position in this evening.

Analysis of Leela for common French variations, approx 1 min per position, or 60k position analysed by latest 30x384 net (which is 4x bigger Neural Network than A0 in 2017)

1. French MuCutcheon, she think white has 58% chance of winning

2. French Winnawer ,---61%

3. French Rubenstein--- 74% which is like pretty hopless for black

For Leela, French is pretty horrible opening choice, she will hardly, (hardly, almost never) play French.

In comparision, Ruylopez is her choice,  55% win chance for white. 

Clearly you do not understand how the device works.  Those numbers are not percent chance of winning.  It's the percent score by White.  They are not the same thing.

What 58 means is that if White were to play that position in 100 games, and White scores 58 percent, that White scores a total of 58 points in those 100 games.

That does not mean they win 58 times out of every 100.  58 points could come in the form of, just as an example, White might win 38 games, 40 draws occur, and Black wins 22 games.  That's just one of many combinations that add up to 58.  That would be a 38 percent win ratio for White, but a total scoring ratio of 58%

It doesn't matter how many draw, the GUI shows by win rate. 

99% win rate may be 99 win 1 loss or 98 win 2 draw but she learn by the same percentile form (0 to 100)  in every step of her learning process. She learn from 25000 games in a single batch and learn those as statistics. Her current learning process doesnt make any difference by the number of draws (or draw rate). In reverse, those statistics are shown back by the GUI as win rate. 

 Alpha Zero develper team think reducing draw rating will make learning process better, but the idea had never been tested in both A0 and current Leela Zero, afaik.

Thank You, for taking the time to do that, I have been working on a extracting McCutcheon games from my messed up database and was curious, I will post stats when I'm done.

I have some suggestions for Leela to be used in World Championship tie break as the substitute for speed chess.

1. As white Leela should move random for the first move.

2.There is feature to set Leela's first move the same move as the above move. It will give the same chance for both candidates.

3. Lela gives the same response for both candidates due to the same moves by both.

4.The rest is following Lela's self response.

The above conditions are to make Leela suitable for The WCC  tie break. 

Does this tells battle between Super GMs especially World Chess Championship, is more of a battle of physical and mental condition? As top GMs have almost equal chess knowledge. Novelties are not really novelties anymore. He who remembers the preparation accurately wins.

Making random the first move is to give no chance for Super GM to prepare all the possible moves by Leela.

Also Leela will move with the depth analysis more than human can do, therefore each candidate might response with different move between both of them.

Very likely both candidates will be lose to Leela. Hence, the winner of both candidates will be determined by their performance throughout the match. The possible criteria to determine the winner is by counting the amount of moves they have made until they lose.

I have compiled a McCutcheon database comprised of 21,754 games.

9245 games are 1-0  42.5%

4655 games are =:=  21.4%

7854 games are 0-1  36.1%

So White takes 53.2% of the points according to my research after 1. e4 e6 2. d4 d5 3. Nc3 Nf6 4. Bg5 Bb4 5. e5 h6.

The databases used were Chessbase 13, Chess Assistant 18, lichess, freechess, and multiple online databases compiled onto a DVD years ago. (containing ICCF, e-mail, engine, and various other sources)

I'm not sure what this data will ultimately do for me, but I have tried to clean and complete headers  and hopefully have minimized doubles. (considering it was hand pruned for 3 months!)

Who gives a d,mn what leela likes or not. Pile of junk.

How strong is Leela on a GTX 1650 gpu? Can't afford laptops with RTX 2060.

They should be pretty close now( Approx -50 elo below 2060 in blitz TC). If 2060 on Leela is 3600 elo, yours should be 3550 elo strength ( in case you correctly configure DX 12 backend in your GPU).

Last month,  Leela developer Ankan, wrote backend for all  dx 12 supported Graphic cards ( It means other cards becomes 4x faster)

My post is mentioned in Leela news channel by main Leela developer, crem, @mooskagh  

Well, according to the finals of CCC 13, Lc0 is stronger than Stockfish since he is winning the match. On the other hand, in the finals of TCEC 17, the duel is tied so far.

On two sites that measure engine strength, http://www.fastgm.de/16-60-0.6.html and https://www.sp-cc.de/nn-longtime-testing.htm, Leelenstein version 14 is the strongest program today, but at CCC 13 it was in third place.