Rating estimate from game

Has anyone built a program to estimate a players rating by examining their moves in a small number of games.
One would expect low rated players to often leaves pieces en pris. And other moves would be indicitaive of higher rated players.
I would imagine that estimating a player's playing strength is a problem for which machine learning could yield interesting results...

The difficulty here is that "normal" chess programs do not play, or approach chess as a whole, in the same way that human players do and would have trouble evaluating moves from a human perspective.

But maybe the newest generation of neural networks might have some prospect of evaluating human play? I don't know.

I think so. You might be right brian.

It's easy enough to estimate ratings (people do it all the time). ;)

There are hundreds of millions of games played on lichess every month.
Each game has recorded ratings for both players.
I would imagine that this would be enough training data for an effective neural network.

@amnonbc said in #5:

There are hundreds of millions of games played on lichess every month.
Each game has recorded ratings for both players.
I would imagine that this would be enough training data for an effective neural network.

I'm actually getting into machine learning myself, and I was thinking of trying something like this. I haven't gotten very far as of now, but I did find a couple other projects:

https://cs230.stanford.edu/projects_fall_2022/reports/19.pdf
https://pranavavva.com/docs/guess-the-elo.pdf

@AsDaGo said in #6:

I'm actually getting into machine learning myself, and I was thinking of trying something like this. I haven't gotten very far as of now, but I did find a couple other projects:

cs230.stanford.edu/projects_fall_2022/reports/19.pdf
pranavavva.com/docs/guess-the-elo.pdf
Well I guess you are right I mean like there are billions of people playing lichess

chess.com actually does this now after each game (starting around a month or two ago).

Accuracy seems to be the biggest factor in the estimated elo given after each game.

@ChefBob said in #8:

chess.com actually does this now after each game (starting around a month or two ago).

Accuracy seems to be the biggest factor in the estimated elo given after each game.

If it makes money, you can be sure chess.com will do it, no mater how ridiculous.

They probably will add a feature of estimating the lifespan of your chessboard by analyzing your playing style

Bro. Chess.com bot vs Chat GPT is the funniest thing to watch. They dont know how to play.