gaia-oc-amd 0.2

Machine learning tool for the determination of new members of open clusters using Gaia data.

Open Cluster Automatic Membership Determination with Gaia data

This repository presents automatic procedure for determining new members of open clusters. We use a neural network architecture to find additional members of an open cluster based on already established probable members. To train our model, we use astrometric and photometric data of stars obtained with the Gaia space telescope. The repository contains code for downloading and preparing datasets, for training the model and for the visualization of some results.

The research paper related to this work can be found here.

Python environment setup

This code is available as a PyPI package and can be installed with

pip install gaia_oc_amd

Alternatively, download the code directly from GitHub with

git clone https://github.com/MGJvanGroeningen/gaia_oc_amd

Data preparation

To use this method, we need to download cluster properties, membership lists and isochrones. To simply get started use:

from gaia_oc_amd import download_startup_data
download_startup_data(save_dir='./data')

This downloads cluster parameters provided by Cantat-Gaudin et al. (2020) and membership lists provided by Cantat-Gaudin et al. (2018), Cantat-Gaudin & Anders (2020), Cantat-Gaudin et al. (2020) and Tarricq et al. (2022). Isochrones are downloaded from the Padova web interface.

In addition, we also want to query the Gaia archive at https://gea.esac.esa.int/archive/ to search for new members, which requires an account for large queries. The code that is used to handle the queries expects a credentials file (with filename 'gaia_credentials' by default), which contains 2 lines for username and password, in the supplied data directory to login to the Gaia archive. This file can be created manually or by running

from gaia_oc_amd import create_gaia_credentials_file
create_gaia_credentials_file(save_path='./gaia_credentials')

which prompts for username and password.

Quick start

With the data and Gaia credentials set up, we can find new members for a cluster with

from gaia_oc_amd import build_sets, evaluate_clusters

cluster_name = 'NGC_1039'

build_sets(cluster_name)
evaluate_clusters(cluster_name)

This performs a cone search on the Gaia archive to obtain source data for potential members and uses a pretrained model included in the package to determine their membership status.

Exploring the code

If you want to explore the methods that are used, be sure to check out the tutorial notebooks in the examples directory. The quick_tutorial.ipynb notebook shows a minimal example while the tutorial.ipynb notebook walks through the methods step by step.

Version 0.2

Major changes

• Improved speed and reduced memory usage of the candidate evaluation by drawing samples with Cholesky decomposition.
• Improved speed of creating deep sets datasets by creating support sets simultaneously.
• The 'fast_mode' for creating features is no longer optional but hard coded, due to mitigation of memory issues.
• Fixed a bug where isochrone data points were inappropriately offset by the estimated extinction effect. These points had effective temperatures outside the range suited for the extinction model. This bug resulted in both a large increase in interpolated isochrone points, which increased feature calculation time, and messed up the zero-uncertainty boundaries in the plots as the grid on which the contour was based was too sparse.
• Only relevant columns are retained when saving source set csv files, reducing storage memory (e.g. columns used to calculate magnitude errors are no longer needed afterwards).
• Option to output the cone data in csv format (which loads faster than votable format, but requires more storage).
• Option to manually set the support set size when evaluating candidates.
• Added a new (default) pretrained model, which was trained longer and on more clusters.

Minor changes

• Changed the iterative process for determining the isochrone and parallax deltas to linearly increase or decrease the threshold value instead of exponentially.
• Added option to set the member fractions, used to determine the isochrone and parallax deltas, in the build_sets function.
• Calculation to the isochrone distance happens in chunks to mitigate memory usage.
• Added scikit-learn to the requirements due to dependence from the dustapprox package.
• Cone search query excludes sources without BP or RP magnitude measurements.
• Fixed bug where the query_vizier_catalog function would fail if no new column names were provided.
• Fixed bug which caused an error when using the modes 'danielski_2018' and 'EDR3' in the correct_for_extinction function.
• Added some new documentation and fixed a number of documentation errors.