...
 
Commits (2)
~~~
otime: 2018-09-21
mtime: 2018-09-21
title: Analysing the 125,262 puzzles in the lichess puzzle set
~~~
*Note: the data used when making this article is probably out of date
by the time you are reading this, but the insights may still be
useful.*
*You can see a larger version of charts below by clicking on them.*
*All code used to fetch, process and plot the puzzle set can be
found on [Gitlab](https://gitlab.com/artefact2/lichess-puzzles).*
# Ratings
[![](/s/img/20180921/plot-ratings-attempts-cnormal-sm.svg)](/s/img/20180921/plot-ratings-attempts-cnormal-lg.svg)
The lichess puzzleset is quite diverse, with ratings spanning from 767
([puzzle #11219](https://lichess.org/training/11219)) from 2926
([puzzle #122656](https://lichess.org/training/122656)). The median
puzzle rating is 1410. 80% of puzzles have a rating between 1040 and
2160.
Player attempts are more biased towards the lower end of the rating
range: 90% of attempted puzzles have a rating below 2000.
[![](/s/img/20180921/plot-votes-avg-sm.svg)](/s/img/20180921/plot-votes-avg-lg.svg)
Puzzles generally receive positive votes by players. However, higher
rated puzzles tend to receive more negative votes. Maybe players vent
out their frustration at the more difficult puzzles they didn't get
correctly by voting them down.
[![](/s/img/20180921/plot-player-ratings-sm.svg)](/s/img/20180921/plot-player-ratings-lg.svg)
Puzzle rating dosen't seem to be strongly correlated with the rating
of the players from the source game.
# Depth
[![](/s/img/20180921/plot-depth-avg-sm.svg)](/s/img/20180921/plot-depth-avg-lg.svg)
As expected, higher rated puzzles require more calculation on average,
but the standard deviation is still very high: harder puzzles are all
over the place, when it comes to calculation depth.
# Time
[![](/s/img/20180921/plot-plies-cnormal-sm.svg)](/s/img/20180921/plot-plies-cnormal-lg.svg)
Tactical opportunities seem to happen regularly throughout the
game. Half of the puzzles happen before the 23rd move. 80% of puzzles
take place between the 11th and the 42nd move.
[![](/s/img/20180921/plot-times-sm.svg)](/s/img/20180921/plot-times-lg.svg)
[![](/s/img/20180921/plot-times-cnormal-sm.svg)](/s/img/20180921/plot-times-cnormal-lg.svg)
The majority of puzzles come from either blitz or rapid games. Puzzles
from bullet games are slightly in the lower end of the rating range,
and correspondence games yield slightly more highly rated puzzles.
# Tactical motifs
*Note: the motifs were detected programatically using
[Ormulogun](https://artefact2.gitlab.io/ormulogun/), which is far
from perfect at identifying everything. Chess is hard.*
[![](/s/img/20180921/plot-motifs-freq-sm.svg)](/s/img/20180921/plot-motifs-freq-lg.svg)
Easier puzzles feature more forks, harder puzzles feature more
discovered attacks and pins. The other motifs seem roughly consistent
across the whole rating range.
[![](/s/img/20180921/plot-motifs-cnormal-sm.svg)](/s/img/20180921/plot-motifs-cnormal-lg.svg)
Puzzles with forks tend to be the easiest, with a median rating of about 1450, while puzzles with
discovered checks or pins tend to be the hardest, with a median rating of about 2200.
# Puzzle types
[![](/s/img/20180921/plot-types-freq-sm.svg)](/s/img/20180921/plot-types-freq-lg.svg)
[![](/s/img/20180921/plot-types-cnormal-sm.svg)](/s/img/20180921/plot-types-cnormal-lg.svg)
Checkmate puzzles tend to be significantly easier than other kinds of
puzzles, and get rarer as puzzle rating increases.
# Endings
[![](/s/img/20180921/plot-endings-cnormal-sm.svg)](/s/img/20180921/plot-endings-cnormal-lg.svg)
[![](/s/img/20180921/plot-endings-freq-sm.svg)](/s/img/20180921/plot-endings-freq-lg.svg)
Ending type seems uncorrelated with puzzle difficulty. Rook endings are
the most common by a fair margin.
\ No newline at end of file
This diff is collapsed.
This diff is collapsed.
This diff is collapsed.
This diff is collapsed.
This diff is collapsed.
This diff is collapsed.
This diff is collapsed.
This diff is collapsed.
This diff is collapsed.
This diff is collapsed.
This diff is collapsed.
This diff is collapsed.
This diff is collapsed.
This diff is collapsed.
This diff is collapsed.
This diff is collapsed.
This diff is collapsed.
This diff is collapsed.
This diff is collapsed.
This diff is collapsed.
This diff is collapsed.
This diff is collapsed.
This diff is collapsed.
This diff is collapsed.
This diff is collapsed.
This diff is collapsed.