Badges
Activity
Ratings Progression
Challenge Categories
Challenges Entered
Learning From Human-Feedback
Latest submissions
Latest submissions
Use an RL agent to build a structure with natural language inputs
Latest submissions
See Allgraded | 198866 | ||
graded | 194590 |
Language assisted Human - AI Collaboration
Latest submissions
Participant | Rating |
---|
Participant | Rating |
---|
NeurIPS 2022 IGLU Challenge
How to parse the world state
Over 1 year agoSo util functions were added to the recent version of gridworld.
Gridworld.tasks
Over 1 year agoHi. Thanks for reaching out. You should be able to install the IGLU gridworld using one of the procedure listed here: GitHub - iglu-contest/gridworld: A reinforcement learning environment for the IGLU 2022 at NeurIPS
Let us know what is the error you are getting, so we can better assist you.
Getting run failures without any logs
Over 1 year ago@dipam can you help with that? I canβt see any logs either.
Some question about data
Over 1 year agoHi, I believe the preferred way to load the data is to use the IGLUDataset
class.
See Dataset section of the starter kit.
@artem_zholus feel free to add anything.
How to save total processes by video
Over 1 year agoHi, you check code examples in the gridworld repo to see how to generate a video. For instance, use the following to try it out
IGLU_RENDER_REALTIME=1 IGLU_HEADLESS=0 python run_env.py
@artem_zholus might have more precision on this but that should get you started.
Computation resource limit in evaluation for IGLU NLP task
Over 1 year agoHi, the resources and limits can be found in the starting kit for the NLP track.
The time allotted for the ranker is 60 minutes and for the classifier it is 30 minutes.
The current resources available are:
- 1 NVIDIA T4 GPU with 16 GB Mem
- 8 vCPUs
- 56GB RAM
How the world state can be parsed/ visualized
Over 1 year ago@artem_zholus can you help answer this question? Thanks.
In the RL Task, can we do pre-training (e.g. imitation learning) with a dataset?
Over 1 year agoYes, those datasets can be used for supervised learning (either as a pretraining step or as an imitation learning approach). We have held out tasks to test for generalization.
Advanced Baseline for RL task NeurIPS 2022 IGLU
Over 1 year agoFor the benefit of others stumbling on this post. Thereβs an active discussion about this issue on our Slack where a potential workaround has been suggested.
How the world state can be parsed/ visualized
Over 1 year agoThe step files correspond to the raw data from the data collection process. For defining the RL tasks in gridworld, we only used information about the initial position/camera of the avatar, the ending state of the world (i.e., target_grid).