This repo contains code for running uct search on atari games, collecting uct search data in form (frame, best_uct_action) and then use it in supervised training.
To run just uct search use:
$ python run_uct.py
To run uct search and collect data for further training use:
$ python collect_uct_data.py
To run supervised training with collected data use:
$ python train_keras.py dirs_with_uct_data
Use flag --help to see all options for UCT and supervised learning scripts.