Skip to content

Latest commit

 

History

History
59 lines (40 loc) · 1.3 KB

README.md

File metadata and controls

59 lines (40 loc) · 1.3 KB
Raw

NeSy-PI

This repository is the official implementation of the paper "Neuro-Symbolic Predicate Invention: Learning Relational Concepts from Visual Scenes"

Setup Locally

  1. Install Python 3.8
  2. Install Pytorch 1.13.1
  3. pip install -r requirements.txt

Setup Remotely

To run the experiments remotely, we provide a docker file Dockerfile in the root folder.

Build docker
docker build -t nesy_pi_docker .
Run docker

Using flag -v to save the dataset folder storage to a perferable place.

docker run --gpus all -it -v path/to/storage:/nesypi/storage --rm nesy_pi_docker

Dataset

The dataset is saved here. Please download the dataset and unzip the folder to NeSy-PI-main\storage.

Experiments

Run Experiments

Check Readme file in folder NeSy-PI/src.

Citation

If you have interest about this work, please consider cite:

@article{nesy_pi,
	author = {Sha, Jingyuan and Shindo, Hikaru and Kersting, Kristian and Dhami, Devendra Singh},
	journal = {Neurosymbolic Artificial Intelligence},
	number = {Preprint},
	pages = {1--26},
	title = {Neuro-symbolic Predicate Invention: Learning relational concepts from visual scenes},
	volume = {Preprint},
	year = {2024}}