RL practice on the path to becoming an AGI researcher (as well as some public proof of life). I'll be implementing algorithms and papers here.
Find me on Twitter @danielpcox if you see anything wrong or otherwise want to chat.
# please read this script first and do something sensible
# virtualenv setup is commented out
./scripts/setup
Train an algorithm with, e.g.:
python main.py train vpg
If you interrupt it with a KeyboardInterrupt exception (Ctrl+C), it'll save the model to /tmp/agent.pt
.
Once you've got a trained agent saved somewhere, you can watch it play Pong with this:
python main.py run /tmp/agent.pt