Cataloging Accreted Stars within Gaia DR2 Using Deep Learning

In [1907.06652], a neural network is used to determine whether a star was accreted onto the Milky Way or was born in situ using only 5D phase space information. The inputs to the network are [l, b, parallax, pmra, pmdec] from the Gaia DR2 catalog. Stars with a score greater than 0.75 are marked as accreted. The subset of stars with 6D phase space information was studied in more detail in [1907.07190] and [1907.07681]. In this release of the catalog, the data is in the HDF5 file format. We recommended using the pandas package within python (with pytables installed). Then the DataFrame can be loaded with <pre><code class="language-python">Stars = pd.read_hdf('Public_AllStars_6D.h5')</code></pre> We have included 4 different files. The scores for all of the stars with \(\delta \varpi / \varpi &lt; 0.1\) are given in Public_AllStars_5D.h5. The subset of these stars which also have line-of-sight velocity measurements are given in Public_AllStars_6D.h5. The files denoted with SelectedByNetwork have only stars which pass the neural network selection as being accreted, with 5D and 6D the same as for AllStars. Each file include the columns <strong>source_id, int64</strong> <strong>l, float64</strong> <strong>b, float64</strong> <strong>ra, float64</strong> <strong>dec, float64</strong> <strong>parallax, float64</strong> <strong>pmra, float64</strong> <strong>pmdec, float64</strong> <strong>phot_g_mean_mag, float32</strong> <strong>phot_bp_mean_mag, float32</strong> <strong>phot_rp_mean_mag, float32</strong> <strong>radial_velocity, float64</strong> <strong>Score, float32</strong> <strong>PhotometricScores, float32</strong> The "source_id" should match with Gaia DR2, allowing for easy cross referencing. The "Score" column is the output of the network using only 5D kinematic information, to be considered as accreted, the score needs to be larger than 0.75. The "PhotometricScores" column is the network output for the neural network which uses the "phot" columns along with the 5D kinematics. The optimal cut for this is 0.9, but as shown in the paper, <em>these results are less robust than the first network</em>. If you use our catalog in any of your works, please cite <br> @article{Ostdiek:2019gnb,<br> author = "Ostdiek, Bryan and Necib, Lina and Cohen, Timothy and<br> Freytsis, Marat and Lisanti, Mariangela and<br> Garrison-Kimmel, Shea and Wetzel, Andrew and Sanderson,<br> Robyn E. and Hopkins, Philip F.",<br> title = "{Cataloging Accreted Stars within Gaia DR2 using Deep<br> Learning}",<br> year = "2019",<br> eprint = "1907.06652",<br> archivePrefix = "arXiv",<br> primaryClass = "astro-ph.GA",<br> SLACcitation = "%%CITATION = ARXIV:1907.06652;%%”<br> }<br> and <br> @article{Necib:2019zka,<br> author = "Necib, Lina and Ostdiek, Bryan and Lisanti, Mariangela<br> and Cohen, Timothy and Freytsis, Marat and<br> Garrison-Kimmel, Shea",<br> title = "{Chasing Accreted Structures within Gaia DR2 using Deep<br> Learning}",<br> year = "2019",<br> eprint = "1907.07681",<br> archivePrefix = "arXiv",<br> primaryClass = "astro-ph.GA",<br> SLACcitation = "%%CITATION = ARXIV:1907.07681;%%”<br> }.<br> Also, please cite <br> @article{Necib:2019zbk,<br> author = "Necib, Lina and Ostdiek, Bryan and Lisanti, Mariangela<br> and Cohen, Timothy and Freytsis, Marat and<br> Garrison-Kimmel, Shea and Hopkins, Philip F. and Wetzel,<br> Andrew and Sanderson, Robyn",<br> title = "{Evidence for a Vast Prograde Stellar Stream in the Solar<br> Vicinity}",<br> year = "2019",<br> eprint = "1907.07190",<br> archivePrefix = "arXiv",<br> primaryClass = "astro-ph.GA",<br> SLACcitation = "%%CITATION = ARXIV:1907.07190;%%"<br> }<br> for any follow up study of Nyx.