Code accompanying "Surveying the space of descriptions of a composite system with machine learning" (2024)

by Kieran A Murphy, Yujing Zhang, and Dani S Bassett [arxiv]

The analyses of the manuscript can be run through train.py and the accompanying command line arguments. For instance, to find the 6 bits of information from a 4x4 Sudoku board with minimal O-information, run the following:

python train.py --dataset_name sudoku --quantity_name oinfo --number_training_steps 20000 --information_in_start 6 --information_in_end 6 --information_in_coefficient_start 1 --information_in_coefficient_end 10

If you want to use the Wikipedia word frequency dataset that we used, download results/enwiki-2023-04-13.txt from https://github.com/IlyaSemenov/wikipedia-word-frequency and put enwiki-2023-04-13.txt into this directory. You can also download another dataset (such as from the Google Web Trillion Word Corpus on Kaggle) and pass its filename through the --ngram_dataset_filename flag.