Tutorial: Machine Learning for Astronomy with Scikit-learn

Download

• Source code: github
• PDF of tutorial: sklearn_tutorial.pdf
• Tarball of exercises and notebooks: exercises.tgz

AstroML

For more information on machine learning for Astronomy, see the astroML code and examples.

Machine Learning for Astronomy with scikit-learn

This tutorial offers a brief introduction to the fields of machine learning and statistical data analysis, and their application to several problems in the field of astronomy. These learning tasks are enabled by the tools available in the open-source package scikit-learn.

scikit-learn is a Python module integrating classic machine learning algorithms in the tightly-knit world of scientific Python packages (numpy, scipy, matplotlib). It aims to provide simple and efficient solutions to learning problems that are accessible to everybody and reusable in various contexts: machine-learning as a versatile tool for science and engineering.

Many of the examples and exercises in this tutorial require the ipython notebook, a tool which provides an intuitive web-based interactive environment for scientific python. Some of the material in the notebooks is duplicated in the following pages, but ipython notebook is required for some parts. For information on how to download the associated notebooks, see the Tutorial Setup and Installation page.

Note

This document is meant to be used with scikit-learn version 0.11+. Find the latest version here.

• 1. Tutorial Setup and Installation
  • 1.1. Python Prerequisites
  • 1.2. Tutorial Files
  • 1.3. Download the datasets
• 2. Machine Learning 101: General Concepts
  • 2.1. Features and feature extraction
  • 2.2. Supervised Learning, Unsupervised Learning, and scikit-learn syntax
  • 2.3. Supervised Learning: model.fit(X, y)
  • 2.4. Unsupervised Learning: model.fit(X)
  • 2.5. Linearly separable data
  • 2.6. Hyperparameters, training set, test set and overfitting
  • 2.7. Key takeaway points
• 3. Machine Learning 102: Practical Advice
  • 3.1. Bias, Variance, Over-fitting, and Under-fitting
  • 3.2. Cross-Validation and Testing
  • 3.3. Learning Curves
  • 3.4. Summary
• 4. Classification: Learning Labels of Astronomical Sources
  • 4.1. Motivation: Why is this Important?
  • 4.2. Star-Quasar Classification: Naive Bayes
• 5. Regression: Photometric Redshifts of Galaxies
  • 5.1. Motivation: Dark Energy, Dark Matter, and the Fate of the Universe
  • 5.2. A Simple Method: Decision Tree Regression
• 6. Dimensionality Reduction of Astronomical Spectra
  • 6.1. SDSS Spectral Data
  • 6.2. Principal Component Analysis
  • 6.3. References
• 7. Exercises: Taking it a step further
  • 7.1. Exercise 1: Photometric Classification with GMM
  • 7.2. Exercise 2: Photometric redshifts with Decision Trees
  • 7.3. Exercise 3: Dimensionality Reduction of Spectra
• 8. Code examples