tools for machine learning and data mining in Astronomy
AstroML: Machine Learning code for Astronomy
AstroML is a Python module for machine learning and data mining built on numpy, scipy, scikit-learn, and matplotlib, and distributed under the 3-Clause BSD license. It contains a growing library of statistical and machine learning routines for analyzing astronomical data in python, loaders for several open astronomical datasets, and a large suite of examples of analyzing and visualizing astronomical datasets.
This project was started in 2012 by Jake VanderPlas to accompany the book Statistics, Data Mining, and Machine Learning in Astronomy by Zeljko Ivezic, Andrew Connolly, Jacob VanderPlas, and Alex Gray.
Core and Addons
The project is split into two components. The core astroML library is written in python only, and is designed to be very easy to install for any users, even those who don’t have a working C or fortran compiler. A companion library, astroML_addons, can be optionally installed for increased performance on certain algorithms. Every algorithm in astroML_addons has a pure python counterpart in the core astroML implementation, but the astroML_addons library contains faster and more efficient implementations in compiled code. Furthermore, if astroML_addons is installed on your system, the core astroML library will import and use the faster routines by default.
The reason for this split is the ease of use for newcomers to Python. If the prerequisites are already installed on your system, the core astroML library can be installed and used on any system with little trouble. The astroML_addons library requires a C compiler, but is also designed to be easy to install for more advanced users. See further discussion in “Development”, below.
This package uses distutils, which is the default way of installing python modules. Before installation, make sure your system meets the prerequisites listed in Dependencies, listed below.
To install the core astroML package in your home directory, use:
pip install astroML
The core package is pure python, so installation should be straightforward on most systems. To install from source, use:
python setup.py install
You can specify an arbitrary directory for installation using:
python setup.py install --prefix='/some/path'
To install system-wide on Linux/Unix systems:
python setup.py build sudo python setup.py install
The astroML_addons package requires a working C/C++ compiler for installation. It can be installed using:
pip install astroML_addons
To install from source, refer to http://github.com/astroML/astroML_addons
There are three levels of dependencies in astroML. Core dependencies are required for the core astroML package. Add-on dependencies are required for the performance astroML_addons. Optional dependencies are required to run some (but not all) of the example scripts. Individual example scripts will list their optional dependencies at the top of the file.
The core astroML package requires the following:
- Python version 2.6.x - 2.7.x (astroML does not yet support python 3.x)
- Numpy >= 1.4
- Scipy >= 0.7
- Scikit-learn >= 0.10
- Matplotlib >= 0.99
- AstroPy > 0.2.5 AstroPy is required to read Flexible Image Transport System (FITS) files, which are used by several datasets.
This configuration matches the Ubuntu 10.04 LTS release from April 2010, with the addition of scikit-learn.
To run unit tests, you will also need nose >= 0.10
The fast code in astroML_addons requires a working C/C++ compiler.
Several of the example scripts require specialized or upgraded packages. These requirements are listed at the top of the particular scripts
- Scipy version 0.11 added a sparse graph submodule. The minimum spanning tree example requires scipy >= 0.11
- PyMC provides a nice interface for Markov-Chain Monte Carlo. Several astroML examples use pyMC for exploration of high-dimensional spaces. The examples were written with pymc version 2.2
- HEALPy provides an interface to the HEALPix pixelization scheme, as well as fast spherical harmonic transforms.
This package is designed to be a repository for well-written astronomy code, and submissions of new routines are encouraged. After installing the version-control system Git, you can check out the latest sources from GitHub using:
git clone git://github.com/astroML/astroML.git
or if you have write privileges:
git clone firstname.lastname@example.org:astroML/astroML.git
We strongly encourage contributions of useful astronomy-related code:
astroML to be a relevant tool for the python/astronomy community,
it will need to grow with the field of research. There are a few
guidelines for contribution:
Any contribution should be done through the github pull request system (for more information, see the help page Code submitted to astroML should conform to a BSD-style license, and follow the PEP8 style guide.
Documentation and Examples
All submitted code should be documented following the Numpy Documentation Guide. This is a unified documentation style used by many packages in the scipy universe.
In addition, it is highly recommended to create example scripts that show the usefulness of the method on an astronomical dataset (preferably making use of the loaders in astroML.datasets). These example scripts are in the examples subdirectory of the main source repository.
We made the decision early-on to separate the core routines from high-performance compiled routines. This is to make sure that installation of the core package is as straightforward as possible (i.e. not requiring a C compiler).
Contributions of efficient compiled code to astroML_addons is encouraged: the availability of efficient implementations of common algorithms in python is one of the strongest features of the python universe. The preferred method of wrapping compiled libraries is to use cython; other options (weave, SWIG, etc.) are harder to build and maintain.
Currently, the policy is that any efficient algorithm included in astroML_addons should have a duplicate python-only implementation in astroML, with code that selects the faster routine if it’s available. (For an example of how this works, see the definition of the lomb_scargle function in astroML/periodogram.py). This policy exists for two reasons:
- it allows novice users to have all the functionality of astroML without requiring the headache of complicated installation steps.
- it serves a didactic purpose: python-only implementations are often easier to read and understand than equivalent implementations in C or cython.
- it enforces the good coding practice of avoiding premature optimization. First make sure the code works (i.e. write it in simple python). Then create an optimized version in the addons.
If this policy proves especially burdensome in the future, it may be revisited.