Overview

Outline

Topics

• Overview of the Bayesian workflow.
• Pointer to resources on generic data science workflows.

Rationale

End-to-end Bayesian analysis contains many steps. This week covers some of the key steps with an emphasis on software tools useful to accomplish them, while building connections with the concepts covered in the class so far.

Bayesian workflows

Some examples of steps involved in effective Bayesian analysis:

• Prior predictive checks: see logistic regression lecture, reinforced in this week’s exercises.
• Validation of MCMC posterior approximation.
• Validation of the model.
• Techniques to identify software defects.
• etc.

On a given problem, depending on the outcome of these steps, different tools will be used, forming a graph of techniques.

Jump to page 5 in Gelman et al., 2020 for a more detailed overview.

General data science workflow resources

We focus in this course on the Bayesian-specific aspects of the data analysis workflow, but these should be combined with generic data analysis workflow practices:

• Organizing your files in a project specific folder.
• Using version control.
• Capturing library dependency versions for reproducibility.
• Follow good coding practice.
• etc.

See, e.g., the software carpentry lessons.