Bayesian Models in Production

Hi Everyone,

Today Ian Costley from Wegmans Food Markets will be giving a talk on how do we get Bayesian models we have worked on in either Stan or PyMC into production once we have finished the initial development, training, and validation cycles.

Here's a short outline of what will be discussed:

• What "production" means in practice (latency, reliability, ownership, monitoring, etc.)
• "Just get it out there deployments" - the best model is a useable model
• Pragmatic / scrappy approaches for serving models on older or available infrastructure
• More structured cloud/MLOps setups (in our case: Azure + Databricks)
• How we use tools like MLFlow and Databricks suite for Stan/PyMC deployments
• Trade offs and general lessons learned