lichess.org
Donate

The xG of Chess: Shark Points

#41
Interesting

But what is the goal of these shark points ?
At football when your team scores 3,2 xG but only one goal, you can conclude that your strikers did a bad job, and this interpretation allows the club to see that their bad results are their striker's fault and then they engage other strikers.

But when it comes to chess I struggle to find a use for these shark points. How they can bring a plus to a player who have these datas ? Only that he can improve mentally so that he converts more winning positions ?
#42
@Testrider said in #30:
> This article has a lot of text, but if I understand correctly the only thing that was actually done was mapping the centipawn eval to a simple curve between 0 and 1?

Hello testrider. Well if that was just this, then at the end of the game you would only get the final score as a value. The point is to map the 'maximum favorable excursion' point of the curve, hence, measure the point at which your chances of winning were maximal. Hope that helps to understand.
#43
@lecw said in #34:
> @BenjiPortheault great post.
>
> What are the "Save" point in the final table ? Are they just the opposite of shark points = how deep in trouble the player once was during the game, yet they managed to draw it ? So Nepo was the best at not-losing (OR, giving winning chances to his opponents, but chances that were impossible to spot)
>
> Seems the Waste and the Save points add up to the same total so I'm guessing it's that.

You are completely correct. This would be also interesting to analyse the 'saving' part. Of course for a single game this is zero sum.
#44
@pseudolol said in #41:
> Interesting
>
> But what is the goal of these shark points ?
> At football when your team scores 3,2 xG but only one goal, you can conclude that your strikers did a bad job, and this interpretation allows the club to see that their bad results are their striker's fault and then they engage other strikers.
>
> But when it comes to chess I struggle to find a use for these shark points. How they can bring a plus to a player who have these datas ? Only that he can improve mentally so that he converts more winning positions ?

thanks pseudolol. I think it would be valuable to know if you are getting good position but won't convert. I guess it's of some value to the player, but likely they already know this from an analysis of their games.
I think for the audience it's good color - player x would be much ahead if he was more clinical in his finishing skills.
#45
@BenjiPortheault said in #42:
> Hello testrider. Well if that was just this, then at the end of the game you would only get the final score as a value. The point is to map the 'maximum favorable excursion' point of the curve, hence, measure the point at which your chances of winning were maximal. Hope that helps to understand.

Centipawn eval (opposed to Average centipawn loss or accuracy) are already determined per move.

I agree that it can be interesting to look at the best, worst and last place in the game (after mapping it to a nice curve) or perhaps the biggest difference. The latter is pretty much how lichess already defined blunders and mistakes.

I just wish that rather than having a lot of talk around the fairly straightforward concept, there would be more around how it is useful (the ending is a bit of an anticlimax).
#46
Well well....an awesome artical but I still can't understand the formula
..I mean for every move will there be an average shark score and we have to average it or something else? Kindly explain this to my dumb brain
#47
@Evo-Bishops said in #46:
> Well well....an awesome artical but I still can't understand the formula
> ..I mean for every move will there be an average shark score and we have to average it or something else? Kindly explain this to my dumb brain

At the end of the (drawn) game, it looks at the one point during the game where one side was better than the other, and computes from there. It is not "for every move" at all.

You could devise another metric which works from an average from all moves or something, if you put further thought into it, and then maybe it's useful or not.
#48
Count the threats that you have compared to your opponent. If it increases in your favor, it's a candidate move. But what if the opponents threats increase faster than your own? Than some might assume their opponent will win the race. Just because the line is in the negative, does not make it a losing game. It's the sway in the slope of the line that points towards who is gaining ground.

The present lichess analysis graph line has 3 directions. Forward (towards a draw), Up or Down (towards a win). After an analysis, use a straight edge like a paper edge and overlap it on the graph. Maintain the left side on the initial move and move the right side of the straight edge up or down as each move changes. Note the peak point of the right edge. Note the FEN code and use that as an initial position. Let the engine start from there and see how the game could have ended. If you need, upload it into a study.

Prune and prune and prune until there is nothing left to prune. That's depth tunnel vision. See the big picture first, the details can come after. Can static positional evaluations be graphed?
#49
@Toscani said in #48:
> Prune and prune and prune until there is nothing left to prune. That's depth tunnel vision. See the big picture first, the details can come after. Can static positional evaluations be graphed?

A fair question! This sounds like a challenge for a tiny chess bot:
youtu.be/iScy18pVR58