Pack Your Model Lunchbox: Model Portability with Pros and Cons

In this talk, we cover the range of methods available to transport, update and replace machine learning models. The goal of model portability is to decrease the difficulty and development time needed to manage models. A side effect of using model objects is the elimination of language bias, by allowing production of language agnostic model objects. Further, model objects can make for easier, more robust CI/CD pipelines. We will discuss the pros and cons of all common model objects, and touch on portability of data models that support such models.
Who is this presentation for?
Machine learning engineers, data scientists, data science management
Prerequisite knowledge:
Data science models and common libraries (xgboost, scikitlearn, tensorflow)
What you'll learn?
The pros and cons of deployment libraries and methods, using schemas to export data models associated with a model object

Anne Lifton has ten years of experience in data science and 3 years in data science management. She has worked across a range of industries from medical devices to retail to engineering, and specializes in reducing the cycle time to delivery of models.