SLIDES - ML Workshop 2: Machine Learning Model Comparison & Evaluation

File uploaded by slimbaltagi on Nov 29, 2017
Version 1Show Document
  • View in full screen mode

A Webinar by Ted Dunning on  How the Rendezvous Architecture Improves Evaluation in the Real World.

 

"Building on the ideas of the key requirements for effective management of machine learning logistics presented in the Overview webinar and in Part I Workshop, we will dive into what model evaluation really can and should be. We will talk about how the rendezvous architecture makes evaluation more effective and also easier. Specifically we’ll cover multi-model comparison, how rendezvous helps you handle metrics, and how it provides query-by-query comparison. A key issue for real world success that is often overlooked by data scientists is latency and system reliability. Conversely, accuracy is often difficult for SysOps team members to address. The rendezvous approach has a built-in way to include latency and accuracy as systematic parts of evaluation, thus addressing key concern of all parts of a DataOps team. Finally, we will discuss how the containerization of models and system components in a rendezvous architecture makes security auditing easier."

Outcomes