Volodymyr wrote:Nicolas,good post.
Nick,very large working done,you believe in this method.
I do not understand why.Just interesting?
Only three games.
Explain the difference. I will repeat, only three games.
My impression is you do not play chess. But you know the rules of the game of chess.
***************************
Only 10 min.per game.What is it?The game of two supercomputers?Perfect game.The masterpiece!
Hi Volodymyr,
I don't know why you want to insult my chess knowledge, its insulting and silly.
The games I showed are just a couple of wins and losses from Tal and Fisher. It shows very well the highs and lows. It shows when they make mistakes and it shows when they play well. The point I was making is that the evaluations show Fischer mistakes and it will show Tal or the mistakes of any other Grandmaster.
I have examples where a 1300 player in a certain game scores as well as Fischer and Tal, therefore I consider this as something that can happen through a game's evaluation, but it doesn't mean a 1300 player will win. That is silly if that is what your posted response is about?
What I am trying to achieve is the average score based on a lot of games played over the table in real matches and not online hence why I have done 250 so far for players around 1400 ELO. Those are factual. Online game I consider flawed as I don't know if the player had his IPhone next to him and has his IPhone playing the moves that he enters into Lichess. Therefore irrelevant and not interesting for the tests that I am doing. As are your examples to argue your point, irrelevant and not interesting.
Of course in any given game a final score could be good or not so good, it depends on the game played by two people. You don't have to question my chess knowledge for that. Common sense is all that is needed for that.
If I wanted to test Tal then you can be assured that I would put several hundred of his games through a thorough test. I don't play around with half assed examples and make assumptions based on them.
You can rest assured before even taking SF8 seriously at 18 ply to evaluate humans I have tested SF8's 18 Ply strength. Even more so because I have evaluated human games in the past 30 to 50 ply deep for other rating tests that I have which are accurate. So of course I would look into SF8 and its accuracy at 18 ply as used by Lichess.
SF8 18 Ply 1.5 - 0.5 Komodo 1.0 (2865 ELO) 40/40
SF8 18 Ply 1.5 - 0.5 Komodo 1.3 JA (2918 ELO) 40/40
SF8 18 Ply 2.5 - 1.5 Hiarcs 14WCSC (2963 ELO) 40/40
SF8 18 Ply 1.0 - 1.0 Critter_1.6a_32bit (3069 ELO) 40/40
================================================== ==
SF8 18 PLY 6.5 - 3.5 Engines
The above chess programs played 1 minute per move against SF8 at 18 Ply. These programs are better than Deep Blue and ran on my modern 2.8 GHz Laptop playing 32 Bit 1 core (exactly the same as the ratings established for the above programs at CCRL 40 40) they search faster and deeper than Deep Blue and overall they searched deeper than SF8's 18 PLY. But SF8 at 18 ply beat the above. Therefore I have every confidence in SF8 18 ply evaluating the mistakes of human accurately as it is better than every human in existence today. There is not a human that exists that thinks 18 ply deep every single move. In summary I don't believe there is a human that would beat SF8 18 ply under real match conditions.
Therefore for the purposes of these tests and exercise which is meant to identify the average strength of players at different strength grades or even to evaluate the average strength of a Tal or Fisher or Carlsen, I consider SF8 more than capable at 18 ply to provide a decent value. Hence the tests.
However, the proof of any tests are in the results. I have not completed the tests for all the different strengths yet and neither have I tested a large set of Grandmaster games.
You can rest assured when tested that I am pretty sure that the GM will show a lot less mistakes than a 1400 player.
![Razz :P](./images/smilies/icon_razz.gif)
But you can also be assured that they will show mistakes as they are human.
Best regards