Differentiating Metal-Poor Main Sequence Stars from White Dwarfs with Spectro-Photometric Observables

Tutorial: Demo Notebook

Read the Paper: arXiv preprint

This repository is an accompaniment to Chandra & Schlaufman (2020, accepted to AJ). The contents are as follows:

• classify_mpms.py is the software deliverable from our paper - a way for anyone to use our logistic regression classifier to differentiate metal-poor main-sequence stars from white dwarfs on the basis of spectro-photometric observables. It includes a convenience function to compute line summaries of the hydrogen Balmer lines, as well as a main classification function that deploys our logistic regression model.

• 0_demo.ipynb is a simple demonstration of the methods described in the paper. It describes how one can use the classes and functions in classify_mpms.py to go from an observed spectrum (and/or photometric observations) to a computed probability of a given star being an MPMS candidate or a white dwarf.

• training_grid.csv contains our grid of training features, derived from the latest synthetic spectra for MPMS and WD stars. For more details, refer to the paper. The classifier trains itself using this grid when initialized. Others are free to use this grid directly to train their own classification algorithm.

• training_grid_dd.csv contains an alternative grid of training features derived from ~ 1500 real stars from SDSS with reliable Gaia parallaxes, and consequently reliable stellar types. We caution that whilst this grid may be preferred by some users for being purely data-driven, it will be quite unreliable for parts of the parameter space that are not covered by the observational data (e.g., very low-temperature stars).

• /paper/ contains all notebooks required to reproduce the plots in the paper, minus large files for the atmospheric model grids. These files can be obtained from the PHOENIX website (MPMS models) and from Simon Blouin (WD models) respectively.

Non-Standard Python Dependancies:

• lmfit
• scikit-learn
• pandas