Anonymous Implementation for ICLR 2026 Submission

This repository contains the implementation of EDISCO, an equivariant continuous-time diffusion model for solving geometric combinatorial optimization problems.

conda env create -f environment.yml
conda activate edisco

EDISCO requires the Cython package for merging the diffusion heatmaps:

cd edisco/utils/cython_merge
python setup.py build_ext --inplace
cd -

Please refer to the data folder.

Please take a look at the reproducing_scripts for details.