• Benchmark Models

Here is an example workflow for a single classifier: Pipeline Demo

We'll want to automate this pipeline process so that we can swap out the following parts for experimentation

• data processing and feature engineering
• classifier (and hyperparameters)
• post processing or ensemble methods

An alternative approach to the typical pointwise prediction is to perform full Bayesian inference. Here is an example implementation of a logistic regression model in pystan: Bradley-Terry Model

With hierarchical Bayesian inference, we can build complex, but interpretable models.