Give Tutor a try

@thibault said in #29:

It does tell you how you fare on your most played openings https://lichess.org/tutor/AsDaGo/blitz/opening

It analyses all games, but not all games go through Stockfish. At the moment a tutor report triggers 100 new stockfish analysis, and also reuses all the analysis that were already available of course.

This is great! Just a though: if the tutor uses all available analysis, could that introduce some selection bias? Someone might only analyse games they thought were interesting, or where they felt they played well (or bad) in. Some might analyse all their lost games, or just won games. If this is not adjusted somehow, this might introduce a bias in the estimates (e.g., the same player will get a more optimistic assessment if they analyse mostly games they won, compared to if they analyse mostly games they lost). Also, some players tend to analyse with an engine more games than others, and this might also skew their results.

Indeed there is room for data bias. Partially countered by the 100 random analysis, but still. Not sure we can avoid it.

I love it!

Nice feature, thanks!
Just one nitpick: it would be better to analyse the past 100 games or so, or at least have an option to do so...

I'm curious how "flagging skills" is composed. A game won by checkmate or resignation due to the opponent blundering in time pressure... well, it's a skillful conversion, even if it's not directly won by flagging the opponent.

@thibault said in #32:

Indeed there is room for data bias. Partially countered by the 100 random analysis, but still. Not sure we can avoid it.

This could possibly be countered by "only" using x games in a given time window, then randomly selecting x games out of this window. For each analysed game that is encountered the amount of games could be encreased.

At first, I had a hard time understanding each stat category. For example: "resourcefulness," seemed vague. I did find what each stat meant after a few clicks, which was good!

Seeing the data is very entertaining, but now as a player, how do I decide what action to take? Perhaps this can be an expansion? "In order to improve this category, you could ...xyz."

It really looks great. Thanks for the hard work!

@thibault said in #1:

Hey fellow lichessers

I'm still working on it, but please do give it a try: https://lichess.org/tutor

And let me know what you think!

Cool but the cooldown kind of sucks.
Mine is for 2h and 13mins

@thibault said in #32:

Indeed there is room for data bias. Partially countered by the 100 random analysis, but still. Not sure we can avoid it.

One option might be to randomly select say 300 games (ideally around 500-1000) and use those as sample, running the analysis when needed. This way, the estimates will be less precise (more random error will be present) but there will be no systematic error. If this was my research project, that's the way I would go, even with a smaller sample: there's no use in having precise but inaccurate measures. Otherwise, it will be mainly a measure of how many and what games people analyse. Thanks for the consideration!

Is there any way to one day add the waiting for lichess tutor chess effects to the home screen. I like the swirly way it shows my games from right to left conveyer belt style and plays moves on them. Would be a cool effect to have at bottom of home page...maybe could be of your last 20 games or something.
Keep up the good work