You might find this FAQ Frequently Asked Questions helpful. In general, I would recommend using Horseshoe instead of Spike and slab, which is recently being studied quite extensively and much easier to work with using HMC. See eg: https://betanalpha.github.io/assets/case_studies/bayes_sparse_regression.html (you can find a pymc3 port here: https://github.com/junpenglao/Planet_Sakaar_Data_Science/blob/master/Ports/Bayes%20Sparse%20Regression.ipynb)
3 Likes