README.md

TQC algorithm

The code includes a modified version of the CleanRL SAC implementation in the file clean_sac_hockey.py, which was used as a baseline comparison.

The final implementation with further info is in tqc.py.

Training

run from se/ folder:

python3 tqc.py --help

Evaluating test runs

run from se/ folder:

tensorboard --logdir runs

Testing models against basic opponents

run from se/ folder:

for file in models/*.pt; do
    python3 hockey_test.py --agent "$file"
done

Testing models against each other

run from se/ folder:

./compare.sh

Tournament connection

run from se/ folder:

./autorestart.sh --args --agent models/models/final_Test_d2_n2_tqc__1__1740091605.pt