PHYS366: Statistical Methods in Astrophysics

Lesson 9: Machine Learning

Goals for this session:

• Use some of the SciKit-Learn utilities to apply some Machine Learning solutions to some simple test problems.
• Describe the difference in approach taken in Machine Learning compared to the statistical inferences more commonly seen in astronomy to date.

Related reading:

• Ivezic Chapters 6, 7, 8 and 9

• Murphy: "Machine Learning: A Probabilistic Approach"

• The SciKit-Learn documentation

Credits:

• The SciKit-Learn tutorial notebooks are based on Andy Mueller's Astro Hack Week 2015 teaching materials.

• The SDSS catalog regression and classification examples are due to Josh Bloom (Astro Hack Week 2014).

See the LICENSE for details about redistributing the SciKit-Learn notebooks.

Package dependency:

• You'll need the SciKit-Learn package installed:

Getting Started in Machine Learning with SciKit-Learn

SciKit-Learn is a well-documented, community-built python package to support data analysis via Machine Learning.

• Let's jump into a SciKit-Learn tutorial, and work through some simple examples.

Applications in Astrophysics

• Having seen some simple examples from the SciKit-Learn tutorial, let's think about how Machine Learning could be applied in astrophysics.

• Here are two example applications from the SDSS catalog example dataset, suggested by Josh Bloom: photometric redshift estimation, and quasar target selection.

Recap

• What should we keep in mind when considering machine learning applications in astrophysics?