Chess.com, still waiting for an explanation please.
Technical question regarding Analysis' Max Analysis
Sort:
Nov 14, 2021
0
#3
I contacted support and did get an answer" to the questions. The conclusions presented in red below.
1. If both are using Stockfish 12 why don't the evaluations match better? Both in moves and in evaluation.
The short answer is that the Cloud Stockfish uses NNUE neural network to classify moves and evaluation of the position. While the browser Stockfish uses the human coded evaluation algorithm.
This explains why Max Analysis typically produces a greater advantage assessment and a different line.
2. The browser is evaluating 5 lines . What is the CLOUD's evaluation line profile?
My interpretation of chess.com's response is just 2 lines not 5.
3. Can an explanation of the resources used for the Cloud max analysis be provided? It clearly runs way quicker then my slowly prodding 5 line browser analysis.
Better hardware an just 2 lines instead of 5
For a NNUE explanation see : https://stockfishchess.org/blog/2020/introducing-nnue-evaluation/
Also some history:
Evolution of a Chess Fish: What is NNUE, anyway?
Stockfish Absorbs NNUE, Claims 100 Elo Point Improvement
Stockfish+NNUE, Strongest Chess Engine Ever, To Compete In CCCC
I would appreciate an explanation of the result difference between running a max analysis (CLOUD) and browser Self Analysis Depth = 99 (Unlimited).
My settings have both using Stockfish 12
Seldom does the Cloud analysis come to the same conclusion as the browser analysis. Often the cloud analysis is more optimistic in its evaluation.
An example to better frame the question:
Given a FEN of
rnbqkb1r/ppp1pp1p/1n4p1/8/3PP3/2N5/PP3PPP/R1BQKBNR w KQkq - 0 6
This an Openings footprint of:
Letting the bowser cook for a good bit to get to depth=30 yields:
Max eval=+1.39, Best browser=+0.79 and they only match for the first move!
Some more time cooking to browser Depth=31 yields:
More time still to depth=32 yields a different 1st move. The initial move match to Max Analysis has moved to 4th of 5.
Simple questions for Chess.com:
1. If both are using Stockfish 12 why don't the evaluations match better? Both in moves and in evaluation.
2. The browser is evaluating 5 lines . What is the CLOUD's evaluation line profile?
3. Can an explanation of the resources used for the Cloud max analysis be provided? It clearly runs way quicker then my slowly prodding 5 line browser analysis.
Thanks in advance.