Simulation-based inference for exoplanetary atmospheric retrieval

This repository is the official implementation of Neural posterior estimation for exoplanetary atmospheric retrieval.

Installation

1. Clone the repository.

   git clone https://github.com/francois-rozet/sbi-ear
   cd sbi-ear
   

2. Download and extract the input data of petitRADTRANS.

   wget https://keeper.mpdl.mpg.de/f/78b3c66857924b5aacdd/?dl=1 -O input_data.tar.gz
   tar -xzf input_data.tar.gz
   

3. Create and activate the conda environment.

   conda env create -f environment.yml
   conda activate ear
   

4. Rebin the opacities to a lower resolution.

   python rebin.py
   

Experiments

Running the experiment scripts requires a Slurm cluster.

1. Generate the training, validation and testing data.

   python generate.py
   

2. When the data generation is finished, launch the training of the estimator.

   python train.py
   

   This step requires to login to Weights & Biases.

3. Run the evaluation notebook.

   jupyter notebook eval.ipynb
   

   It is necessary to modify the runpath according to the run name.