"Zoobot" Galaxy Zoo Bayesian CNN and Active Learning - Public Release

This repository contains code to simulate training a Bayesian CNN using active learning on Galaxy Zoo, for the paper:

Galaxy Zoo: Probabilistic Morphology through Bayesian CNNs and Active Learning, M. Walmsley et al, Monthly Notices of the Royal Society, 2019 (submitted)

This code is released for reproducibility only. The code is an exact copy of the code used for the paper. As a research best practice, the code is documented, tested, and (we hope) fairly readable. However, we strongly suggest any users interested in further original research wait for the launch of Yarin Gal's Bayesian Deep Learning Competition (Q3 2019), which will include much of this code in a friendlier form.

Installation

The core features are provided as the zoobot package. Install as per Python convention: pip install -r requirements.txt pip install -e zoobot

Usage

We provide example scripts for using the zoobot package in the root directory.

• run_zoobot_on_panoptes.py trains and evaluates a model on a train/test split of either GZ2 or GZ DECALS galaxies. This is useful for developing models.
• panoptes_to_tfrecord.py and gz2_to_tfrecord.py convert the respective download catalogs into tfrecords suitable for models.

To simulate active learning, see the readme in zoobot/active_learning.

Note that the Galaxy Zoo 2 and Nasa Sloan Atlas galaxy catalogs required are available externally from here and here.

Zoobot Folders

• estimators includes custom TensorFlow models (and associated training routines and input utilities) for galaxy classification.
• active_learning applies those models and routines in an active learning loop. Currently, previous classifications are used as an oracle. See the Readme in this folder for more details.
• get_catalogs is used to download GZ2 classifications. Panoptes classifications have been refactored to the repo gz-panoptes-reduction.
• tests contains unit tests for the package. Look here for examples!
• tfrecord has various useful utilities for reading and writing galaxy catalogs to tfrecords.
• uncertainty has code to find the coverage fractions of trained models

Contact Us

This code was written by Mike Walmsley. Please get in touch by email.

Legal Stuff

Copyright (C) 2019 - Mike Walmsley

This program is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version.

This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.

You should have received a copy of the GNU General Public License along with this program. If not, see http://www.gnu.org/licenses/.