Hidden Markov models in PyMC3.
Distribution
classes for HMM state sequences (DiscreteMarkovChain
) and mixtures that are driven by them (SwitchingProcess
)FFBSStep
) that works with NUTS—or any other PyMC3 samplerTransMatConjugateStep
)Distribution
classesTo use these distributions and step methods in your PyMC3 models, simply import them from the pymc3_hmm
package.
See the examples directory for demonstrations of the aforementioned features. You can also use Binder to run the examples yourself.
Currently, the package can be installed via pip
directly from GitHub
$ pip install git+https://github.com/AmpersandTV/pymc3-hmm
First, pull in the source from GitHub:
$ git clone git@github.com:AmpersandTV/pymc3-hmm.git
Next, you can run make conda
or make venv
to set up a virtual environment.
Once your virtual environment is set up, install the project, its dependencies, and the pre-commit
hooks:
$ pip install -r requirements.txt
$ pre-commit install --install-hooks
After making changes, be sure to run make black
in order to automatically format the code and then make check
to run the linters and tests.