Inferring interior structures of exoplanets with Mixture Density Networks

---

❗ An improved version of this model is available at https://github.com/philippbaumeister/ExoMDN ❗

---

This repository contains the trained machine learning models and python notebooks for the paper Machine-learning inference of the interior structure of low-mass exoplanets (Baumeister et al. 2020).

Required packages

This project requires Python 3.

• keras = 2.2.4
• numpy = 1.18.0
• scipy >= 1.2.0
• matplotlib >= 3.0.2
• tensorflow = 1.15.2
• tensorflow-probability = 0.7.0
• ipywidgets >= 7.4.2
• joblib >= 0.13.2
• scikit-learn = 0.22.1

Installing the required packages

Using anaconda (preferred)

conda env create -f requirements.yml

Activate with

conda activate tf1.15

Using pip

pip install -r requirements.txt

How to use

• MDN_exoplanets.ipynb contains all the code to load the trained MDN models and predict the distribution of possible interior structures of a planet.
• The mdn directory contains the MDN layer code adopted from https://github.com/cpmpercussion/keras-mdn-layer.
• The models directory contains data scalers and the MDN models trained either with mass and radius of the planet, or with mass, radius, and k₂.