Is 100 accuracy even possible for chess engines?

TeacherOfPain

Well accuracy shouldn't be determined on what Engines prefer, but rather what is played in a game. It is very interesting how engines do this...

chamo2074
pinkblueecho a écrit :
chamo2074 hat geschrieben:

d4 and e4 are the moves that the engine doesn't favour, it prefers 1.e4 and 3.Nf6 or Nc6

 

Sorry, there seems to be a typo there with regards to e4. Could you clarify that?

I meant the engine prefers Nf6 or Nc6 in terms of evaluation

chamo2074
TeacherOfPain a écrit :

Well accuracy shouldn't be determined on what Engines prefer, but rather what is played in a game. It is very interesting how engines do this...

Wdym?

pinkblueecho
chamo2074 hat geschrieben:

d4 and e4 are the moves that the engine doesn't favour, it prefers 1.e4 and 3.Nf6 or Nc6

 

Strange. I just checked this on Stockfish 11 at depth 48. It gives d4, c4 and Nf3 +0.2 each. It doesn´t like e4 (-0.1!) very much and, in fact, prefers e3 (0.0)!

TeacherOfPain

@chamo2074

IDK, it just feels like sometimes engines have innaccuracies within them, like for example a good move would be seen as an innaccuracy, or a excellent move would be seen as just a good move, etc.

I just feel like it is dependent on the algorithm and either what the engine prefers or what it is trained to see to get higher accuracies, but in this way I don't feel it is correct, and should be changed by people who maintenize and build the chess engines. Perhaps it is a lot of work and is a small tweek, but knowing full accuracy would be a nice tribute rather than knowing 99% of it. 

But I guess that is just my perspective on the matter.

chamo2074

Yes, I can give an e.g, if engine can do something without sacrificing, the sacrifice that leads to the same thing is an innacuracy

Ripley_Osbourne

By that standart, checkmating by promoting to queen should be an innacuracy compared to checkmate by promoting to rook...

Ripley_Osbourne

... Or reversed? surprise.png

chamo2074
Ripley_Osbourne a écrit :

By that standart, checkmating by promoting to queen should be an innacuracy compared to checkmate by promoting to rook...

Well, the color code usually says that both are alternatives, the engine sometimes underpromotes to a rook when it leads to the same, no idea why. Same when it repeats moves and then play the correct move, I have no idea why they do it

Ripley_Osbourne

The underpromotion responds to the idea of "economy of means", that surely have been implemented in their code. Repeating moves allows to calculate longer for the same one move that will be played in the end. When humans do this, it's usually in order to regain time by increment, or try reach 40° move without losing on time, or check if the opponent wants a draw or not, and possibly force them to play an inferior move if they want the win at all costs.

captainnegi

no,,, 

chess1gamess

Can anyone program 10x10 chess with two additional pieces Generals (it is like an upgraded Knight) who moves like a king but covering 2 squares instead of 1 (i.e. covering not 9 squares, but 25) and that are situated between king/queen and bishop. Pawns moves 1,2,3 squares from initial position and en passant rule applies. All other rules are the same, just castling will be longer by 1 square (3 squares instead of 2).

Ripley_Osbourne
chess1gamess a écrit :

Can anyone program 10x10 chess with two additional pieces Generals (it is like an upgraded Knight) who moves like a king but covering 2 squares instead of 1 (i.e. covering not 9 squares, but 25) and that are situated between king/queen and bishop. Pawns moves 1,2,3 squares from initial position and en passant rule applies. All other rules are the same, just castling will be longer by 1 square (3 squares instead of 2).

 

I see you're no newbie, why are you doing such off topic post here instead of opening your own thread about it?

MananDhyani

https://www.chess.com/a/2HXP8S8SS2GEcn

 

chess1gamess
Ripley_Osbourne wrote:
chess1gamess a écrit :

Can anyone program 10x10 chess with two additional pieces Generals (it is like an upgraded Knight) who moves like a king but covering 2 squares instead of 1 (i.e. covering not 9 squares, but 25) and that are situated between king/queen and bishop. Pawns moves 1,2,3 squares from initial position and en passant rule applies. All other rules are the same, just castling will be longer by 1 square (3 squares instead of 2).

 

I see you're no newbie, why are you doing such off topic post here instead of opening your own thread about it?

by mistake. I was posting (copy-paste) to another place

TeacherOfPain

Yeah, it just seems sometimes that the engines have preferences @chamo2074, we may never know, but regardless it is something to take note.