CosmoFlow

This is the repository that goes along with the paper CosmoFlow: Scale-Aware Representation Learning for Cosmology with Flow Matching

We implement a flow-matching based model for representation learning of cold dark matter density fields.

1. Clone the repository:

   git clone https://github.com/sidk2/cosmo-compression.git
   cd cosmo-compression

2. Install dependencies:

   pip install .

• Training: See src/cosmo_compression/train.py for training script.
• Experiments: Explore the examples/notebooks/ directory for Jupyter notebooks demonstrating model usage, interpolation, and downstream tasks.
• Pretrained Models: Pretrained checkpoints are available in the models/ directory.
• Data: We include a subset of the 1P dataset from CAMELS to facilitate running the examples.

Open a notebook in examples/notebooks/ and follow the instructions in the cells to reproduce experiments and visualizations.

If you use this codebase in your research, please cite the corresponding paper (citation to be added).

For questions or contributions, please open an issue or contact skannan@ucsb.edu.