I'm using ChessBase, and ok Perfect Gent - that makes a lot more sense now. Thank you.
Rybka analysis?
Sort:
Question, PerfectGent.. You said the minus sign is in blacks favor. Is that global? Or would it be more correct to say the minus sign means it's in your opponents favor? And a plus sign is in your favor?
For most engines (including rybka), minus sign always means black is in favor.
ok. But then ... let me take this example.
I'm playing white, computer is playing black. It's analysing as black to move and it's coming up with +.52 and such on it's hundredths of a pawn. Now - if I decide to ask it to analyse the positing for MY move (which I do at times) now it's coming up with -.57 and so on. As if the meaning of the sign has swapped.
Let me copy/pase two such lines.
for white to move:
14 -0.63 485119 0:04.52 Nc2 Ra6 Ne3 Bb5 Nc2 Qb7 Bf4 e5 Bg5 Bd7 Be7 Qb3 Raa1
white moves Nc2 - for black to move:
15 +0.56 1.8M 0:13.69 Ra6 Nb4 Raa8 Nc2 Rc8 Bf4 e5 Bg5 f6 Bh4 Qc6 f3
I don't really care about what it means when it's my turn and such - I just want to make sure I'm reading what it's saying correctly.
Ok. I will try that. Thank you.
It is not "analyzing as black". It is just "analyzing". The analysis processes are separate from the move selection processes.
Rybka analyses the current position, neither from white or blacks perspective but from the boards. The advantage is given as a positive number if white is at advantage, and a negative number for black. Then the move selection process asks the analysis process to analyze each candidate move 10 positions deep (or whatever), and selects the one that is the best move. If it is playing as black, it will select the one that is rated with the lowest number. If it is playing as white, it will select the one that has the highest number.
The reason that the numbers changed is that they are comparing different positions. the move Nc2 gave white a significant positional advantage which Rybka rated as .99 pawns.
Ok. Then I have another follow up question. When it's White's turn (in the current game I'm playing/referring to), it displays the number as negative, and then when it's blacks turn, it displays the number as positive. So if that's the case, the evaluation number it's giving would be for the position the board is in and NOT the relative strength of the suggested move for the line?
"white moves Nc2 - for black to move:
15 +0.56 1.8M 0:13.69 Ra6 Nb4 Raa8 Nc2 Rc8 Bf4 e5 Bg5 f6 Bh4 Qc6 f3"
So this would mean that the board currently is at about a half pawn advantage for white - NOT that this series of moves evaluates to a half pawn advantage for white?
I *did* find this on the Rybka Forum FAQ.
"Essentially, the evaluation score Rybka displays should reflect her confidence in winning the game, assuming that the contempt value properly reflects the strength of the opponent."
Yes, that is the score of the analysis of the board, NOT the analysis of the principal variation.
Ok then - one more question (I'm annoying, aren't I):
Why does the evaluation end up slightly different between the lines? O.o (some are +.62 .. some are +.58 .. etc etc) Or would that be something only a Rybka programmer would be able to answer?
Interesting that Fritz's analysis gives the evaluation (score) that results from the variation shown NOT the analysis of the board. Wonder why?
Also:
Note that the score takes into account material imbalances, e.g.:
If Black has lost a Knight and White a pawn: Black's turn to move and Fritz analysis is:
12... a6 13. h4 h6 14. g4 Re8 15. Bg2 Nd7 16. g5 h5 17. O-O 2.34/10.
The Knight (3.0) minus the pawn (1.0) = 2.0 then White’s first move advantage (0.3) = 2.3 so the score means the variation ends with no great positional value.
The_Joy_of_Rooking wrote:
Yes, that is the score of the analysis of the board, NOT the analysis of the principal variation.
To answer your question JoishiBoidi, what this guy wrote is wrong. The score for a certain variation rybka gives is the evaluation of the board at the END of the variation. Note this is not necessarily only the sequence of moves given by rybka in the variation, sometimes it gives only two moves of a variation, while it is actually looking 12 ply deep.
Ok. Thank you very much!
Sothilde wrote:
The_Joy_of_Rooking wrote:
Yes, that is the score of the analysis of the board, NOT the analysis of the principal variation. To answer your question JoishiBoidi, what this guy wrote is wrong. The score for a certain variation rybka gives is the evaluation of the board at the END of the variation. Note this is not necessarily only the sequence of moves given by rybka in the variation, sometimes it gives only two moves of a variation, while it is actually looking 12 ply deep.
But the analysis of the principle variation IS the analysis of the position. It scores the position based on (what it thinks is) perfect play for both sides 12 ply deep.
Yes, if a position is even in material, space, etc, and there is a forced tactical gamebreaker in a couple of moves, the score will be heavily weighted towards that side.
If an engine didn't take future moves into account in position scoring, it wouldn't work very well =).
The_Joy_of_Rooking wrote:
Sothilde wrote:
The_Joy_of_Rooking wrote:
Yes, that is the score of the analysis of the board, NOT the analysis of the principal variation. To answer your question JoishiBoidi, what this guy wrote is wrong. The score for a certain variation rybka gives is the evaluation of the board at the END of the variation. Note this is not necessarily only the sequence of moves given by rybka in the variation, sometimes it gives only two moves of a variation, while it is actually looking 12 ply deep. But the analysis of the principle variation IS the analysis of the position. It scores the position based on (what it thinks is) perfect play for both sides 12 ply deep. Yes, if a position is even in material, space, etc, and there is a forced tactical gamebreaker in a couple of moves, the score will be heavily weighted towards that side. If an engine didn't take future moves into account in position scoring, it wouldn't work very well =).
I'm sorry but you're really wrong here. The score an engine gives to a position is based on it's evaluationg criteria besides a material count, for example (just hypothetical values) +.35 when you have the bishop pair in an open position, +0.1 when your rook is on an open file etc. This is just for one position, not for a variation.
When you let it analyze a position to get the best move, simply said, it makes every possible move, and gives a score to the board resulting after each move. Then it makes the possible counter moves (skipping obviously wrong moves and emphasizing the logical moves using some complicated algorithms). Then it again gives a score to the board resulting after each counter move. In that way it examines a lot of moves (the node count) and picks after each iteration the moves with the highest score for white and the lowest score for black, resulting in the presented principle variation.
So the score present with each variation is the score AFTER the variation is played, not the score of the current position. In a sense you are right the score after the best variation is the evaluation of the position, but technically it's not and more important it's not what he was asking for.
Here's a final interesting comment:
I finally got around to playing a full game vs the computer today. I was white and the computer was black. I got checkmated (naturally, since I'm nowhere near GM status) and the final analysis the computer made for it's checkmate move listed the hundredths of pawns value as +99.99 ... that would sort of imply that + is black's advantage and - is white's?
Oct 13, 2008
0
#17
You are all wrong about evaluation signum. Plus/minus sign before the evaluation score depends mainly upon GUI's translation of engine otuput. There is ABSOLUTE score ({+} is in whites favour and {-} is in blacks favour - this is what you all are talking about and this is what most GUIs and engines give as output). But there is also RELATIVE score (where {+} means advantage for side on the move, and {-} means advantage for other side of the board). So if there is no change in balance on the board and after half move engine shows change in sign this means it is showing RELATIVE SCORE.
JoishiBoido's GUI obviously shows relative evaluation, that explains why it evaluates black mating white as +99.99.
This is not a BUG as diskamyl pointed out!
For example if you are using Winboard or WinboardX for analysis it shows relative evaluation score (+ means advantage for side on the move):
http://www.chess.com/download/view/portable-winboardx--rybka-22---updated-setup
I downloaded the Rybka chess engine off the site here to play against the computer some and see about improving my strategy a little bit by looking at what the computer is thinking. However, I need a little bit of help deciphering what it's displaying, and the .txt files that came with it didn't make sense or seem to address what I wanted to know.
So - here are a few questions:
What does the Evaluation Diagram display mean? What is the +5 and -5 on the graph supposed to be a measurement of?
And also for the Engine Output, let's take an example line:
5 -0.13 3455 0:00.05 Nf6
What's the first number shown mean? (the 5) I think the 2nd number is how it will effect the Evaluation Diagram graph, but that doesn't help me much since I don't know what the Evaluation Diagram is a measure of? I think the 3455 is some measurement of time that it took to decide on that move, but I'm not sure about it. Obviously the 0:00.05 is how many minutes/seconds it took thinking for the following line of moves.
Anyway - any help explaining or deciphering some of this stuff would be appreciated. Or if this question has already been posted and re-posted, a simple link will probably be enough.