PyMC3 is a python module for Bayesian statistical modeling and model fitting which focuses on advanced Markov chain Monte Carlo fitting algorithms. Its flexibility and extensibility make it applicable to a large suite of problems.
Check out the Tutorial!
PyMC3 is Beta software. Users should consider using PyMC 2 repository.
- Intuitive model specification syntax, for example,
x ~ N(0,1)translates tox = Normal(0,1) - Powerful sampling algorithms, such as the No U-Turn Sampler, allow complex models with thousands of parameters with little specialized knowledge of fitting algorithms.
- Easy optimization for finding the maximum a posteriori(MAP) point
- Theano features
- Numpy broadcasting and advanced indexing
- Linear algebra operators
- Computation optimization and dynamic C compilation
- Simple extensibility
- Transparent support for missing value imputation
- PyMC3 Tutorial
- PyMC3 paper
- Bayesian Modelling in Python -- tutorials on Bayesian statistics and PyMC3 as Jupyter Notebooks by Mark Dregan
- Coal Mining Disasters model in PyMC 2 and PyMC 3
- Global Health Metrics & Evaluation model case study for GHME 2013
- Stochastic Volatility model
- Several blog posts on linear regression
- Talk at PyData NYC 2013 on PyMC3
- PyMC3 port of the models presented in the book "Doing Bayesian Data Analysis" by John Kruschke
- The PyMC3 examples folder
- Manual of PyMC3 distributions
The latest version of PyMC3 can be installed from the master branch using pip:
pip install --process-dependency-links git+https://github.com/pymc-devs/pymc3
The --process-dependency-links flag ensures that the developmental branch of Theano, which PyMC3 requires, is installed. If a recent developmental version of Theano has been installed with another method, this flag can be dropped.
Another option is to clone the repository and install PyMC3 using python setup.py install or python setup.py develop.
Note: Running pip install pymc will install PyMC 2.3, not PyMC3, from PyPI.
PyMC3 is tested on Python 2.7 and 3.3 and depends on Theano, NumPy, SciPy, Pandas, and Matplotlib (see setup.py for version information).
In addtion to the above dependencies, the GLM submodule relies on Patsy.
scikits.sparse enables sparse scaling matrices which are useful for large problems. Installation on Ubuntu is easy:
sudo apt-get install libsuitesparse-dev
pip install git+https://github.com/njsmith/scikits-sparse.git
On Mac OS X you can install libsuitesparse 4.2.1 via homebrew (see http://brew.sh/ to install homebrew), manually add a link so the include files are where scikits-sparse expects them, and then install scikits-sparse:
brew install suite-sparse
ln -s /usr/local/Cellar/suite-sparse/4.2.1/include/ /usr/local/include/suitesparse
pip install git+https://github.com/njsmith/scikits-sparse.git
Salvatier J, Wiecki TV, Fonnesbeck C. (2016) Probabilistic programming in Python using PyMC3. PeerJ Computer Science 2:e55 https://doi.org/10.7717/peerj-cs.55
See the GitHub contributor page