We use convolutional neural networks (CNNs or convnets) to predict galaxy properties using Sloan Digital Sky Survey (SDSS) gri images. Gas-phase metallicity, which is often estimated by using optical spectroscopy, can also be estimated using our CNN.
We describe our methods in a paper: Wu & Boada (2018).
Download this repository by running
git clone https://github.com/jwuphysics/galaxy-cnns.git
cd galaxy-cnns
All analysis was performed inside the Jupyter notebooks using a Python3 environment. We use the verion 0.7.0 of the fastai machine learning framework built atop Pytorch. This can be installed by running the following:
git clone https://github.com/fastai/fastai.git
cd fastai
conda env create -f environment.yml
Note that you will need a GPU. If you don't have one, substitute the previous last line with this instead:
conda env create -f environment-cpu.yml
Before executing any code (or running any notebooks), enter the environment
by running conda activate fastai
(or conda activate fastai-cpu
).
If you encounter any errors, please feel free to reach out to me (@jwuphysics) or check this post on the fastai forums.
We queried the SDSS DR14 image cutout service
using the script ./download_images.py
in order to obtain gri images.
We queried the SDSS MPA-JHU DR7 catalog
of spectral line and derived galaxy properties using the commands in the SQL
script,
./SDSS_sql_query.sql
.
To run the our notebooks, make sure that you are in the fastai
conda environment first,
and then run jupyter notebook
and enter the ./notebook
directory.
If you wish to reproduce all figures from the paper, run the notebooks in the
./notebook/paper
directory. You will first need to have downloaded all of the
data and executed the notebooks labeled
"06. Predicting stellar mass in addition to metallicity.ipynb" and
"10. The effects of resolution.ipynb" first (sorry about this disorganization -- this
may be cleaned up in a future update).
If you would like to reference our paper, please use the following citation, produced by NASA ADS:
@ARTICLE{2018arXiv181012913W,
author = {Wu, J.~F. and Boada, S.},
title = "{Using convolutional neural networks to predict galaxy metallicity from three-color images}",
journal = {ArXiv e-prints},
archivePrefix = "arXiv",
eprint = {1810.12913},
keywords = {Astrophysics - Astrophysics of Galaxies},
year = {2018},
month = {oct}
}