manutej / new_world_disco

A working repo where we are iterating on working CNN's that discover exoplanets

new_world_disco

A working repo where we are iterating on working CNN's that discover exoplanets. This New World Discovery repository should be forked by anyone interested in playing with Convolutional Neural Networks and the lightkurve.org python library to learn about and find new exoplanets.

Background

This work is built on the PyTorch code created by Dr. Megan Ansdell et al at NASA's Frontier Development Lab in 2018 (link to gitlab code repo)

Her work was an improvement upon the prior work of Dr. Andrew Vanderburg and Chris Shallue et all from 2016 (link to github code repo)

Datasets

The dataset to start out with if you are using the exonet.py program can be found here. You should download and unzip these files in a nearby directory for easy use.

You should use Lightkurve.org to pull new data for this project. As time goes on, there will be more tutorial-like resources availble to help you preprocess the data, but in gerneral, see the papers from Ansdell et al and Vanderburg et al cited below as they walk you through how to prepare your data pipeline for model training and inference.

Cite if you use this code

If you want to use this code in your work, please cite in the comments of your code where it came from in case anyone else stumbles upon your repo, they should know where it came from. If you plan to use this code in any scholastic or professional work or wish to publish, you must fully cite the following papers and sources:

This work has been put together by the efforts of the Data Science and Engineering Teams at SpaceLab.space with the inspiration, guidence, and foundational work provided by:

Scientific Domain Knowledge Improves Exoplanet Transit Classification with Deep Learning, Ansdell et al, -The Astronomical Journal Letters

Identifying Exoplanets with Deep Learning: A Five-planet Resonant Chain around Kepler-80 and an Eighth Planet around Kepler-90, Vanderburg et al, -The Astronomical Journal

Identifying Exoplanets with Deep Learning. II. Two New Super-Earths Uncovered by a Neural Network in K2 Data, Anne Datillo & Andrew Vanderburg et al, -The Astronomical Journal

A Technique for Extracting Highly Precise Photometry for the Two-Wheeled Kepler Mission, Vanderburg and Johnson, -Publications of the Astronomical Society of the Pacific