Position Evaluation now working differently?
Sort:
Hi there,
it seems like the recent update a couple of days ago the evaluation of the current position in the analysis has changed. Now the value of the current position is always exactly the value of the best next move. Until a couple of days ago, that was not the case. I guess it made sense that there was a difference between these two values, because you cannot assume that you or your opponent play the best move 100 % of the time.
Anyway: Why did this change and is it going to stay this way? Or am I missing something?
Regards,
Kornspitz123
I'd always found it awkward that they were different, because my naive expectation was that they should be the same. After a bit of thought, I'd decided that the difference was due to technical details about how positions and lines were scored by the various engines. Just a guess because I've not seen a precise definition of eval on chess.com.
Aside, I've noticed that up until very recently I was able to run Stockfish 15 in Chrome but today I just noticed that it's been limited to Stockfish 11. This probably has an impact on evaluations.
Slightly technical reference:
https://blogs.cornell.edu/info2040/2022/09/30/game-theory-how-stockfish-mastered-chess/#:~:text=Stockfish%20uses%20a%20static%20evaluation,grandmasters%20for%20over%20a%20decade.
Thanks a lot for your input!
I just noticed that what I mentioned above is by far not the biggest problem with analysis.
Just had a game where in the analysis he valued my 27. ... Re4 as +1.00 because that is the value of whites best next move 28. Bc3. However, when I click on Bc3, he shows the value as -0.73. The values don't change when I give him more time to calculate, or when I go forth to 28. Bc3 and then back to 27. ... Re4.
Had a couple of issues like this.
I guess it must have something to do with server limitations given the current situation. But unfortunately, the way it is the analysis is pretty useless. :-(
I hope it gets better again.
Regards,
Kornspitz123
Engines employ certain tricks to speed up computation, including "pruning" certain low potential lines and caching the scores for previously calculated lines/positions. This has the plus of allowing the engine to return scores fairly quickly but it also has a couple of significant negatives. In particular, scores might change if you exit and rerun analysis (...they're not deterministic) and scores might not update if you explore a line and return to an earlier point, since the scores may have been cached.
Note that if you dig into a line by, say, one move, then the eval will subsequently look one move further than it previously did, and the score will change. The score may revert if you pop back up to an earlier position since it may use a previously cached value.
Please Sign Up to comment.
If you need help, please contact our Help and Support team.