Bayesian statistics allows you to build models that respect and exploit domain-specific knowledge while also properly accounting for uncertainties. Physics is an example of this – where often the models we are trying to fit have a huge parameter space, but not all combinations of parameters “make sense”. One often solves this by introducing regularisation but the bayesian approach makes this idea more general. Stan is a probabilistic modelling language that is suitable for Bayesian modelling and PyStan is its Python interface.
In this webinar, Faculty Data Scientist Omar Sosa Rodriguez will discuss:
- An introduction to bayesian statistics
- How bayesian statistics differs from traditional machine learning
- How to implement a simple bayesian logistic regression using Stan
- Learning as inference — The bayesian perspective on traditional machine learning
- Bayesian inference with Stan
This talk is aimed at data scientists who have some experience with machine learning and Python, but no previous knowledge of Bayesian statistics or Stan is assumed.