The original data was downloaded from https://database.lichess.org/. We used the data from January and Februrary of 2015.
The preprocessed pickle file can be found here. This file is already stripped
of all non-evaluated games and unnecessary metadata, and the gameplay has been parsed into a list of moves and a list of blunders. When using the .pkl files instead of the raw data, line 111 in main.py can be commented out. This reduces the runtime of the program significantly, but still takes around 45 minutes. When using smaller PGN files (e.g. November+December 2013) the total runtime is around 5-10 minutes.
The output of the scripts are four matplotlib graphics, showing different aspects of the players blunder behaviour. One of the four is the heapmat showing the blunder activity across the board:
