Lichess Analysis!
I analyzed high-level rated blitz games from Lichess and built heatmaps for three questions:
where checks happen most, where captures happen most, and where white pawn promotions happen most.
Lichess is open and publishes game data through the Lichess database,
which makes projects like this possible.
Dataset Slice
Rated Blitz games from 2024-01 to 2024-05.
Games Processed
201,791,204 PGNs.
Compute
153 GB compressed (~1.09 TB uncompressed), processed in 38 minutes with C.
For the visualizations below, I filtered to stronger games (both players above 2000
ELO).
Riskiest Square
Given white is in check, on which square is the white king most likely to be?
Cell values are percentages.
Deadliest Square
If a white piece is captured, where is that capture most likely to happen?
Cell values are percentages.
Promotion Column
When white promotes a pawn, which file sees promotions most often?
Full code for data processing and plotting is on GitHub.