Dear all, I am pleased to announce Autologistic.jl, a package for autologistic regression (ALR) modelling. The ALR model is an extension of logistic regression. It’s a probabilistic graphical model for binary responses (including covariate effects).
Some important features of the package:
- Analyze ALR models with arbitrary graphs.
- Estimation and inference using maximum likelihood for small models, or pseudolikelihood + parametric bootstrap for larger models.
- Includes different variants of the ALR model (change the numeric coding of the responses, or use different forms of centering).
- Extensible design to allow different parametrizations of the model.
- Draw random samples from the models using Gibbs sampling or several different perfect sampling implementations.
Fairly extensive documentation gives more detail and examples. Deeper technical background can be found in the article Better Autologistic Regression.
This is my first serious project in Julia, after many years working with R and MATLAB. I have to say the experience of developing in Julia has been very positive, especially considering that Julia is still a “new” language. And in the end my samplers run well over 100x faster than they did in MATLAB. So hats off to everyone in the Julia community, I’m a convert.