This is the repository for the paper "Sourcerer: Sample-based Maximum Entropy Source Distribution Estimation".
Install all dependencies with pip install -e .
hydra is used for configuration management and command line interface.
You can run all benchmark tasks either from the command line or interactively (with VSCode cells or Jupyter notebooks) using the benchmark_simulator_script.
To run them interactively, overwrite the local_overrides list with your desired configuration.
Alternatively, you can run the script from the command line: python3 benchmark_simulator_script.py simulator=two_moons (as an example, the two_moons task is selected here).
By default, results will be saved in results_sourcerer.
To run the Hodgkin-Huxley experiment, the public dataset from Scala et al. (2020) is required. The dataset can be downloaded from https://dandiarchive.org/dandiset/000008/draft.
Alternatively, a preprocessed dataset along with a set of simulations to train the surrogate is publicly available at https://github.com/berenslab/hh_sbi.
To train the surrogate, use the hh_sims_and_stats script.
To perform source estimation after training the surrogate, use the hh_script.
Both can be run interactively using VSCode cells or Jupyter notebooks.
Code for the figures can be found in the figures folder.
To reproduce the figures, it is necessary to first run the experiments and update the paths to point to the result files.
Parts of this repository are based on code from the following public repositories:
https://github.com/berenslab/hh_sbihttps://github.com/MaximeVandegar/NEB
@inproceedings{
vetter2024sourcerer,
title={Sourcerer: Sample-based Maximum Entropy Source Distribution Estimation},
author={Julius Vetter and Guy Moss and Cornelius Schr{"o}der and Richard Gao and Jakob H. Macke},
booktitle={The Thirty-eighth Annual Conference on Neural Information Processing Systems},
year={2024},
url={https://openreview.net/forum?id=0cgDDa4OFr}
}