Delighted to present the NetHack Learning Dataset (NLD) at #NeurIPS2022 next week!
NLD is a new large-scale dataset for NetHack and MiniHack, aimed at supercharging research in offline RL, learning from observations, and imitation learning.
1/
NLD is a new large-scale dataset for NetHack and MiniHack, aimed at supercharging research in offline RL, learning from observations, and imitation learning.
1/
NLD has 3 parts. First, NLD-NAO, a dataset of 10 billion state transitions from 1.5 million games recorded on the alt.org/nethack server from 2009-20.
That's bigger than MineDojo in trajectories, with unprecedented coverage of NetHack, including 22,000 ascensions!
2/
That's bigger than MineDojo in trajectories, with unprecedented coverage of NetHack, including 22,000 ascensions!
2/
Next up, NLD-AA, 3 billion state-action-score transitions from 100,000 trajectories recorded by the winning bot of the NetHack Challenge - AutoAscend.
This symbolic bot trounced the best neural agents in the challenge by 3x, but is still very short of full ascension.
3/
This symbolic bot trounced the best neural agents in the challenge by 3x, but is still very short of full ascension.
3/
Finally, NLD provides code to allow users to create, load and stream their own trajectory datasets, and to do so highly performantly!
Just like @NetHack_LE , we want users without large budgets to be able to do large-scale research! We're faster than SCII & original MineRL!
4/
Just like @NetHack_LE , we want users without large budgets to be able to do large-scale research! We're faster than SCII & original MineRL!
4/
NLD code ships with NLE version 0.9.0, and the instructions to download the data is on the NLE README.
We also have a colab you can play around with!
Paper: arxiv.org/abs/2211.00539
Colab: colab.research.google.com/drive/1GRP15Sb…
5/
We also have a colab you can play around with!
Paper: arxiv.org/abs/2211.00539
Colab: colab.research.google.com/drive/1GRP15Sb…
5/
Finally, I'm lucky to have had such great collaborators @robertarail, @DanielleRotherm, Vegard Mella, @_rockt, @HeinrichKuttler, @NailaMurray!
And thanks to @dtype at NAO + Maciej & Michal Sypetkowski (AutoAscend)!
To learn more, drop by Hall J on Thurs at 4CST, or DM!
6/6
And thanks to @dtype at NAO + Maciej & Michal Sypetkowski (AutoAscend)!
To learn more, drop by Hall J on Thurs at 4CST, or DM!
6/6
• • •
Missing some Tweet in this thread? You can try to
force a refresh
