NeurIPS 2021: MineRL Competition. Intro track baseline submission kit.

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.

How to submit.

1. Clone this repository.
2. Update aicrowd.json file with your list of authors.
3. 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!

Contents

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.

1. Updated aicrowd.json to specify intro track with "tags": "intro". Also set "gpu": true so that GPU is used for running the model.
2. 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!).
3. Added the PPO model potato.zip to ./train directory.
4. Updated submission_test_code.py by placing functions from the baseline code into the file, and updating the main entry point inside run_agent_on_episode (at the end of the code file).