https://github.com/brendel-group/compositional-ood-generalization
Official code for the paper "Compositional Generalization from First Principles" (NeurIPS 2023)
https://github.com/brendel-group/compositional-ood-generalization
Science Score: 10.0%
This score indicates how likely this project is to be science-related based on various indicators:
-
○CITATION.cff file
-
○codemeta.json file
-
○.zenodo.json file
-
○DOI references
-
✓Academic publication links
Links to: arxiv.org -
○Academic email domains
-
○Institutional organization owner
-
○JOSS paper metadata
-
○Scientific vocabulary similarity
Low similarity (6.2%) to scientific vocabulary
Repository
Official code for the paper "Compositional Generalization from First Principles" (NeurIPS 2023)
Basic Info
Statistics
- Stars: 7
- Watchers: 1
- Forks: 3
- Open Issues: 0
- Releases: 0
Metadata Files
README.md
Compositional Generalization from First Principles
This repository provides the codes for all experiments shown in the paper Compositional Generalization from First Principles.
Setup
- use Python 3.9
- run
pip install -r requirements.txt - if desired, a Dockerfile is provided. You can create the container with
docker build .
Running the code
- all experiments where run on a single NVIDIA-RTX-2080Ti
- config files for all experiments can be found in
cfgs/ - to start an experiment, run
main.py --config path/to/experiment/config.yml
Citation
If you find the insights from the paper or our code base useful, please cite
@misc{wiedemer2023compositional,
title={Compositional Generalization from First Principles},
author={Thaddäus Wiedemer and Prasanna Mayilvahanan and Matthias Bethge and Wieland Brendel},
year={2023},
eprint={2307.05596},
archivePrefix={arXiv},
primaryClass={cs.LG}
}
Owner
- Name: brendel-group
- Login: brendel-group
- Kind: organization
- Repositories: 9
- Profile: https://github.com/brendel-group
GitHub Events
Total
- Watch event: 2
Last Year
- Watch event: 2
Dependencies
- python 3.9-buster build
- matplotlib *
- numpy *
- pandas *
- seaborn *
- spriteworld *
- torch *
- torchmetrics *
- torchvision *
- tqdm *
- wandb *