exoplanet is a toolkit for probabilistic modeling of time series data in astronomy with a focus on observations of exoplanets, using PyMC3. PyMC3 is a flexible and high-performance model building language and inference engine that scales well to problems with a large number of parameters. exoplanet extends PyMC3’s language to support many of the custom functions and distributions required when fitting exoplanet datasets. These features include:
A fast and robust solver for Kepler’s equation.
Scalable Gaussian Processes using celerite.
Fast and accurate limb darkened light curves using starry.
Common reparameterizations for exoplanet-specific parameters like limb darkening and eccentricity.
And many others!
All of these functions and distributions include methods for efficiently calculating their gradients so that they can be used with gradient-based inference methods like Hamiltonian Monte Carlo, No U-Turns Sampling, and variational inference. These methods tend to be more robust than the methods more commonly used in astronomy (like ensemble samplers and nested sampling) especially when the model has more than a few parameters. For many exoplanet applications, exoplanet (the code) can improve the typical performance by orders of magnitude.
exoplanet is being actively developed in a public repository on GitHub so if you have any trouble, open an issue there.
Where to find what you need
🖥 For general installation and basic usage, continue scrolling to the table of contents below.
🖼 For more in depth examples of exoplanet used for more realistic problems, go to the Case studies page.
📈 For more information about scalable Gaussian Processes in PyMC3 (this was previously implemented as part of exoplanet), see the celerite2 documentation page.
👉 For helper functions and PyMC3 extras that used to be implemented as part of exoplanet, see the pymc3-ext project.
Copyright 2018, 2019, 2020, 2021 Daniel Foreman-Mackey.
The source code is made available under the terms of the MIT license.
If you make use of this code, please cite this package and its dependencies. You can find more information about how and what to cite in the citation documentation.
These docs were made using Sphinx and the Typlog theme. They are built and hosted on Read the Docs.
Adds high-level, pure-Theano/Aesara implementation of RegularGridInterpolator
to replace compiled version #167
Moves compiled Ops to exoplanet-core making exoplanet a pure-Python package #171
Moves ReboundOp and ReboundOrbit to rebound-pymc3 #166
Removes deprecated submodules (gp
, optimize
, and sampling
) and base distributions (e.g. xo.Angle
) #165
Adds support for PyMC3 v3.10 and Theano-pymc
Fixes issue with automatic execution of tutorials
Adds faster solver for Kepler’s equation
Improves automation of documentation and release process
Removes most C++ Ops and replaces them with pre-compiled functions which allows the distribution of binary wheels
Removes gp
submodule; moved to new celerite2 package
Deprecates some non-exoplanet-specific functions and distributions; moved to pymc3-ext package
Fixes compatibility with PyMC3 version 3.9
Fixes a documentation bug introduced in v0.3.1
Adds support for light travel time when computing positions in Keplerian orbits
Adds SecondaryEclipseLightCurve
for modeling eclipsing binaries
Adds UnitDisk
distribution for fitting eccentricity vectors
Adds tests and support for Windows
Adds a “Jacobian” interface to the orbit objects for reparameterization
Adds support for fitting circular orbits with duration
Adds a bls_estimator
for transit search using Astropy’s BoxLeastSquares
Improves infrastructure for generating documentation
Adds an EclipsingBinaryLightCurve
for building binary star models
Adds DensityDist
implementation for celerite GP likelihoods
Makes rebound
an optional dependency
Adds ConditionalMeanOp
and DotLOp
for scalable conditional mean calculation
and prior sampling with celerite
Adds developer documentation
Moves documentation to a separate repository
Adds TTVOrbit
tutorial
Switches tutorials to lightkurve for data access
Improves packaging and code style features
Fixes bugs and improves interface to TTVOrbit
Adds a new interface for tuning dense mass matrices with less overhead
Adds support for photodynamics using rebound
Adds a new interface for assigning units to Theano variables
Adds new physically-motivated distributions for impact parameter and eccentricity
Improves test coverage
Fixes bug in diagonal elements of the IntegratedTerm
model
Fixes bug in indexing for TTVOrbit
models with large TTVs
Updates starry
to get much better performance for high order spherical
harmonics
Renames StarryLightCurve
to LimbDarkLightCurve
Restructures the C++ backend to reduce code duplication
Adds support for fitting of astrometric observations
Adds support for exposure time integration in celerite
models
Adds new distributions for periodic parameters and U(0, 1).
Fixes many small bugs
Fixes some edge case failures for Kepler solver
Improves reliability of contact point solver and fails (more) gracefully when this doesn’t work; this reduces the number of divergences when fitting a transit model
Improves contact point solver using companion matrix to solve quadratic
Improves reliability of Angle
distribution when the value of the angle
is well constrained
Improves the reliability of the PyMC3Sampler
Adds a new optimize
function since the find_MAP
method
in PyMC3 is deprecated
Adds cronjob script for automatically updating tutorials.
Adds a more robust and faster Kepler solver (ref)
Fixes minor behavioral bugs in PyMC3 sampler wrapper
Adds regular grid interpolation Op for Theano
Fixes major bug in handling of the stellar radius for transits
Fixes small bugs in packaging and installation
Fixes handling of diagonal covariances in PyMC3Sampler
Initial public release