Machine Learning Models and Productionization

Despite growing teams and budgets, very few machine learning models reach the production stage. This session discussed:

  1. What is model productionization? What makes it difficult?
  2. How does it relate to MLOps?
  3. How should one plan for production ML?
  4. What kinds of systems and process should be in place for ensuring successful delivery?
  5. What does the team structure and skillset look like?

Panelists:

  1. Aditya Patel, Director, Data Science of Glance (InMobi; Bangalore)
  2. Nischal HP @nischalhp, VP of Technology, Omnius (Berlin).

Takeaways

  1. Uncertainty cannot be eliminated in the performance of the model. So extensive testing, circuit breakers, fallback mechanisms, and incremental deployment mechanisms are critical.
  2. ML development and deployment should be thought about within the context of the product development.
  3. Many of the systems, methods and ideas are applicable here including versioning but need to be repurposed to suit the context

Ask questions, post comments for the speakers and hosts on the Comments section.

Previous session: The first session was held on 20 May. Summary of the session is available here: https://hasgeek.com/fifthelephant/making-data-science-work-1/

About the curators: Venkata Pingali and Indrayudh Ghoshal of Scribble Data have curated this session. Scribble Data is a Bangalore/Toronto startup, active in the data community.

About the series producer: The Fifth Elephant is platform for practitioners working with data (engineering, to application of data science for different use cases) to showcase their work and to collaborate.

For further inquiries, contact 7676332020 or write to fifthelephant.editorial@hasgeek.com

Videos

See all
Model Productionization

Model Productionization

Aditya Patel, Nischal H P. Moderated by Venkata Pingali, Indrayudh Ghoshal

1 hour 3 June 2020

Hosted by

Jump starting better data engineering and AI futures