A pytorch solution for 4th place NIPS 2019 MineRL Competition
pip install -r requirements.txt
Once you install this requirements, you can run our code. This solution did not use human datasets.
Run minerl environments without a head(servers without monitors) use a software renderer such as xvfb：
xvfb-run -s '-screen 0 1024x768x24' python3 train.py
Run mineral environments with a head(servers without monitors attached):
python3 train.py
The model will be saved in this this folder train/ when training complete. The training process takes 72h(1x2080Ti + 4xCPU) .
Run it just like training:
xvfb-run -s '-screen 0 1024x768x24' python3 test.py
python3 test.py
The average reward of test is between 30 and 40 after normal training.
- Docs: http://www.minerl.io/docs/
- Github: https://github.com/minerllabs/minerl
- AIcrowd: https://www.aicrowd.com/challenges/neurips-2019-minerl-competition
- Competition Proposal: https://arxiv.org/abs/1904.10079
- Human datasets: https://router.sneakywines.me/minerl_v1/data_texture_0_low_res.tar.gz
- Extra human datasets：https://router.sneakywines.me/minerl-v123321123321/data_texture_0_low_res.tar.gz