This is the repository for the preprint 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 either via the commandline or interactively (with VSCode cells or jupyter notebooks) using benchmark_simulator_script.py.

To run the interactively, overwrite the local_overrides list with your desired configuration.

Alternatively, launch the script in the command line python3 benchmark_simulator_script.py simulator=two_moons with your desired configuration (as an example, the two_moons task is selected here).

By default, the Inverse Kinematics task with a differentiable simulator will be performed and 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 at https://dandiarchive.org/dandiset/000008/draft.

Alternatively, a preprocessed dataset, together with a set of simulations to train the surrogate is publically available at https://github.com/berenslab/hh_sbi.

To train the surrogate, use the hh_sims_and_stats.py script. To perform source estimation after training the surrogate, use the hh_script.py script. Both scripts can be run interactively using VSCode cells.

Code for the figures can be found in the figures folder. To reproduce the figures, it is required to first run the experiments and update the paths to point to the result files.

The code is mainly based on the pytorch library. Configuration management is performed with hydra.

Parts of the code are based on code from the following publically available repositories:

• https://github.com/berenslab/hh_sbi
• https://github.com/MaximeVandegar/NEB