Abstract: This session is a hands on workshop (with coding) to demonstrate how to gain observability (monitoring & alerting) for production machine learning pipelines. We will provide background on why observability is important to run successful MLOps, then walk through in detail how to set up a robust observability system.
Without a proper observability system, it is impossible to scale a successful machine learning effort. The session will provide ML engineering teams with the tools they need (all available in the open source ecosystem) to solve major visibility gaps in the machine learning lifecycle, including monitoring data quality, job statuses, ML model performance, and retraining.
The session will cover the end-to-end process, from data prep jobs running in Airflow, to model development and experimentation in Jupyter notebooks, to model serving in production.
The content covered will be of interest to data engineers and data scientists, including anyone who is working on machine learning projects.
We recommend that participants have strong backgrounds in python and at least high level knowledge of job orchestrators like Airflow, which are used to run automated data pipelines.
Bio: Coming Soon!