How good are titled players? Lichess data analysis.

tldr: They are very good.

A while ago, I downloaded the profile information of over 300,000 Lichess players. Of those, 384 have "confirmed" titles in their profiles.

Here's a plot of the ratings of each of those players. Note that I eliminated players with fewer than 20 games played in a category, and that this was before the Classical/Rapid split.

imgur.com/a/BHguG

very nice.

Nice graphs. Perhaps visualizing the data as Bell curves would make it a bit easier to see the densities at each rating and compare the average ratings of players with different titles.

@thijscom

Good idea! Here you go:

imgur.com/a/YSVKl

Notice how the GM Classical density is to the left of the IM's. That's because they don't really play all that many classical games.

@dudeski_robinson Thanks, that looks very nice!

And indeed, it's kind of interesting that for bullet and blitz, in both cases we have the rough ordering GM > IM > FM > LM > NM > CM (as roughly expected), while for classical the ordering appears to be IM > GM > LM > FM > NM > CM. So it's not just GMs and IMs being in the "wrong" order, but also FM and LM swapping places. Maybe I should start playing classical again and boost the FM curve to the right :)

hehe.

yeah. The main problem is that I use a low threshold: Only 20 games played. Higher thresholds would be more reliable, but the dataset would be too small.

How did you go to download the 384 confirmed titles profiles ?
I would like to know them.

@Catastrov I downloaded the profiles of 300,000 lichess members. Among those profiles, 384 have "confirmed" titles in their profiles.

How have you downloaded the profiles of the 300 000 lichess members ?

Thanks.

@Catastrov I downloaded a list of games here:

database.lichess.org

I extracted the list of all players in those games. Then, I used the Lichess API documented here with a Python script to extract profile information:

github.com/ornicar/lila#http-api