11-09, 15:30–16:15 (America/New_York), Music Box (5th floor)
There is currently a chasm between training models in Python and building a production prediction service around that model. It is assumed that you need systems experience just to build an end-to-end ML system to showcase your model to stakeholders.
In this talk, we will show you how to build a serverless ML system using only a workflow engine (Github Actions), a platform to manage your features and models (Hopsworks), and a UI (Streamlit). The complete system will be written in only Python and run on serverless services.
The system we will introduce is an energy price prediction service. We will engineer features in Pandas, use Hopsworks' Python centric API to register features and use features them for training and serving. We will detail the advantages of a Python-based domain specific language (DSL) for point-in-time correct feature selection and skew-free transformations over the alternative SQL approach. Finally we are going build a dashboard making real time predictions using Streamlit
No previous knowledge expected
Fabio Buso is VP of Engineering at Hopsworks, leading the Feature Store development team. Fabio holds a master’s degree in Cloud Computing and Services with a focus on data intensive applications.