mssm 1.2.0

Toolbox for estimating Generalized Additive Mixed Models (GAMMs), Generalized Additive Mixed Models of Location Scale and Shape (GAMMLSS), and more general smooth models.

mssm: Mixed Sparse Smooth Models

Description

[!NOTE] Our preprint detailing the algorithms implemented in the mssm toolbox is now available on arXiv.

mssm is a toolbox to estimate Generalized Additive Mixed Models (GAMMs), Generalized Additive Mixed Models of Location Scale and Shape (GAMMLSS), and more general (mixed) smooth models in the sense defined by Wood, Pya, & Säfken (2016). Approximate estimation (and automatic regularization) of the latter only requires users to provide the (gradient of) the log-likelihood. Furthermore, mssm is an excellent choice for the modeling of multi-level time-series data, often estimating additive models with separate smooths for thousands of levels in a couple of minutes. mssm also supports fully Bayesian inference about model coefficients and regularization parameters.

Note: The main branch is updated frequently to reflect new developments. The stable branch reflects the latest releases. If you don't need the newest functionality, you should install from the stable branch (see below for instructions). Documentation (built from the stable branch) is hosted here - together with a tutorial for mssm! Plotting code to visualize and validate mssm models is provided in this repository!

Installation

The easiest option is to install from pypi via pip. This can be achieved in two steps:

1. Setup a conda environment with python > 3.10
2. Install mssm via pip

The latest release of mssm can be installed from pypi. So to complete both steps (after installing conda - see here for instructions), simply run:

conda create -n mssm_env python=3.13
conda activate mssm_env
pip install "mssm[plot,mcmc]"

Note: This will also install optional dependencies, listed in the square brackets, required for model visualization, parallelization of selected functions, and convergence statistics for the mcmc sampler. Also note, that pypi will only reflect releases (Basically, the state of the stable branch). If you urgently need a feature currently only available on the main branch, consider building from source.

Building from source

You can also build directly from source. This requires an installation of eigen. setup.py first checks whether eigen has been installed via conda from conda-forge, alternatively checks for eigen in "usr/local/include/eigen3", and finally falls back to cloning it from their git (this will fail if you have not set up git on your machine). After cloning and navigating into the downloaded mssm repository you can then install via:

pip install .

Contributing

Contributions are welcome! Feel free to open issues or make pull-requests to main.