Skip to content

Installation Guide (MacOS)

Jump to bottom
rpgoldman edited this page Jun 5, 2021 · 44 revisions

PyMC3 installation on MacOS

We recommend using Anaconda to install Python3 on MacOS, which allows for packages like PyMC3 to be installed using its conda utility. If you are running on the new M1 chip, this will run PyMC3 (and Python) using Rosetta2, rather than natively on Apple silicon. We have yet to be able to easily install PyMC3 on Python running natively with the M1 chip, due to some problems building dependencies.

The latest release of PyMC3 can be installed from Conda-Forge:

conda install -c conda-forge mkl pymc3 theano-pymc

Note that you must specifically request theano-pymc or you will get an obsolete version of PyMC3 that works with the now-abandoned theano library. We encourage you to test this with the --dry-run flag to ensure you get up-to-date versions.

You may find it better to use conda create to create a new virtual environment for your work with PyMC3.

While strongly discouraged due to installation problems you could try to install PyMC3 and its dependencies via PyPI using pip:

pip install theano-pymc
pip install pymc3

The reason installation via PyPI is difficult, especially on Windows and OSX, is that Theano requires compilation against MKL, which is difficult to set up, while Conda comes with its own compilers and MKL installation.

Note that running pip install pymc will install PyMC 2.3, not PyMC3, from PyPI.

The current development branch of PyMC3 can be installed from GitHub, also using pip:

pip install git+https://github.com/pymc-devs/pymc3

Make sure to install the Theano-PyMC version that is pegged in PyMC3's requirements.txt. For more information, see the Theano-PyMC installation section, below.

Lastly, another option is to clone the repository and install PyMC3:

git clone https://github.com/pymc-devs/pymc3
cd pymc3
python setup.py install (or python setup.py develop)

Theano-PyMC Installation

PyMC3 is tested on Python 3.6, 3.7, and 3.8 and depends on NumPy, SciPy, and pandas (see requirements.txt for version information).

Moreover, as of late 2020, PyMC3 now requires Theano-PyMC instead of Theano.

  • If you have Theano installed, please remove it and install Theano-PyMC via Conda Forge:
conda remove theano
  • If you have an outdated version of Theano-PyMC, then re-install the module via Conda Forge:
conda remove theano-pymc -y
conda install -c conda-forge theano-pymc -y
  • If you have cloned PyMC3's repository onto your computer, you can install Theano-PyMC using the requirements.txt:
pip install -r requirements.txt

Arviz Installation

Plots and diagnostics are entirely done by ArviZ - if you follow the installation instructions above, then it will be installed alongside PyMC3.

Optional

In addtion to the above dependencies, the GLM submodule relies on Patsy.

Clone this wiki locally