When Grandmasters Blunder

data-analysischessmapreducedistributed-computing

Key points:

Used Pachyderm (distributed file system + MapReduce engine) to analyze 4,899,067 chess moves from 2014 games
Crafty chess engine annotation mode quantifies blunders in pawn units; a 2-pawn deficit is nearly always a loss at top level
Only 1.37% of all moves were 2-pawn blunders; for players rated above 2775 (Carlsen/Anand level) the rate drops to 0.96%
Carlsen-Anand double blunder in WCC game 6 was a ~1-in-10,000 pair-of-moves event; across a 12-game match (~600 pairs) roughly 1-in-20
Blunder rate vs. rating fits an exponential: gaining 600 rating points halves blunder frequency
Processing ran on Google Compute Engine and took ~6 hours; full run limited Crafty to 2 seconds analysis per move

Abstraction: MapReduce analysis of chess blunder rates across rating levels

Connections: Pachyderm · Magnus Carlsen · Mapreduce · Data Analysis