Before running, data should be generated into a data/ folder, and, if using a pretrained model, that model should be placed into a models/ folder.
To use the trained model mentioned in the paper, use model_32_4.pt, which is in results.
The following python files hold central classes and functions of AlphaTensorCS184:
architecture.py training.py selfplay.py tensorgame.py training.py utilities.py
Additionally, main.py holds the three primary modes in which you may want to run the loop at-large.
Prior to running a training method, you should generate data with example_generation. The samples of data are too large to store, and relatively fast to generate.
If you are running this program, the following additional folders are used:
models/ data/
Within tests/, there are three scripts used for testing. They should be put into the primary folder before running:
tests.py testing.ipynb selfplay_tests.py
analysis.ipynb is used to generate information for the report, based on information within results.