NimSed / astro-machines

Machines Learn to Infer Stellar Parameters Just by Looking at a Large Number of Spectra

Astro-Machines:

Machines Learn to Infer Stellar Parameters Just by Looking at a Large Number of Spectra

Here you can find the code and (pointers to) the data used in the paper: arxiv.org/abs/2009.12872 (DOI:10.1093/mnras/staa3540)

Please refer to the main project page, to try out the 'sliders' and 'RETR-SPECT' interfaces.

How to use

• Clone the repository to a local directory of your choice. e.g. astro-machines.

  • git clone https://github.com/NimSed/astro-machines.git
  • cd astro-machines

• Extract the lists, metadata and labels for HARPS spectra:

  • cd lists
  • tar xfz list.tar.gz
  • cd ..

• Play with the provided notebook to visualize the learned features and finally reproduce the plots in section 6 of the paper.

  • cd notebooks
  • jupyter-notebook

Optional -- to pass individual HARPS-like spectra through pretrained networks

• Fetch pretrained models:

  • cd models
  • sh fetch_models.sh
  • cd ..

• Test/infer on the provided sample.fits:

  • cd infer
  • python3 infer.py

You can of course use an arbitrarily chosen HARPS spectrum (in fits format). Just pass it to infer.py using the --fits_file argument.

Dependencies

pip3 install astropy