astroNN is a python package to do various kinds of neural networks with targeted application in astronomy by using Keras API as model and training prototyping, but at the same time take advantage of Tensorflow or PyTorch flexibility.
For non-astronomy applications, astroNN contains custom loss functions and layers which are compatible with Keras v3. The custom loss functions mostly designed to deal with incomplete labels. astroNN contains demo for implementing Bayesian Neural Net with Dropout Variational Inference in which you can get reasonable uncertainty estimation and other neural nets.
For astronomy applications, astroNN contains some tools to deal with APOGEE, Gaia and LAMOST data. astroNN is mainly designed to apply neural nets on APOGEE spectra analysis and predicting luminosity from spectra using data from Gaia parallax with reasonable uncertainty from Bayesian Neural Net. Generally, astroNN can handle 2D and 2D colored images too. Currently astroNN is a python package being developed by the main author to facilitate his research project on deep learning application in stellar and galactic astronomy using SDSS APOGEE, Gaia and LAMOST data.
For learning purpose, astroNN includes a deep learning toy dataset for astronomer - Galaxy10 Dataset.
Uncertainty Analysis of Neural Nets with Variational Methods
- Henry Leung - Initial work and developer - henryskyAstronomy Student, University of TorontoContact Henry: henrysky.leung [at] utoronto.ca
- Jo Bovy - Project Supervisor - jobovyAstronomy Professor, University of Toronto
This project is licensed under the MIT License - see the LICENSE file for details