This repository contains the baseline solution for the intro track of NeurIPS MineRL 2021 competition, placed inside the submission kit. This means it is ready for a submit with few changes!
Other Resources
- Repository for the baseline solution - Original baseline solution in a cleaner format. Go here if you want to study the code without overhead from the competition kit!
- Submission template - The original submission template with a random agent. Go here if you want full details on the submission process.
- Clone this repository.
- Update
aicrowd.json
file with your list of authors. - Follow the instructions here to submit to AICrowd.
Alternatively, see this video for a step-by-step guide that shows how to do the submission fully online, without using the command line or having to download/install anything on your local machine!
This kit contains the "RL plus script" baseline solution, modified to fit into the submission baseline. See the original script on how to train the model, the code here only runs a pretrained agent.
Here is a list things that were modified over the submission template to get things working.
- Updated
aicrowd.json
to specify intro track with"tags": "intro"
. Also set"gpu": true
so that GPU is used for running the model. - Updated
environment.yml
with the correct Python, PyTorch and stable_baselines3 versions (note: it is important that you make sure these versions match your local setup, otherwise the agent may not work!). - Added the PPO model
potato.zip
to./train
directory. - Updated
submission_test_code.py
by placing functions from the baseline code into the file, and updating the main entry point insiderun_agent_on_episode
(at the end of the code file).