Project Repository for Nethack Challenge .
AI611: Deep Reinforcement Learning, KAIST GSAI 2022SP
First, run the docker container:
$ ./docker/run.sh
Then, install additional dependencies within the docker container:
$ python3 -m pip install stable_baselines tensorboard
(Note that this step is needed, since we do not build a custom docker image,
but use the image provided by fairnle/challenge:dev
.)
Afterwards, you can freely train our agent:
$ python3 train_pop_art_agent.py
$ tensorboard --logdir /tmp/nethack --host 0.0.0.0 --port 6006
Afterwards, open localhost:6006
on your browser to see the
logged outputs.
After training the agent, you can record the video as follows:
$ python3 record.video.py ${CKPT_FILE} # e.g. '/tmp/nethack/run-022/log/nh-pa-last.pt'
$ python3 -m nle.scripts.ttyplay /tmp/record/nle.PPPPP.I.ttyrec.bz2
Where PPPPP
is the process ID and I
is the environment index.
In general, you'd want to set PPPPP
to the latest one, and I = 0
.