A recent post by Mehdi Mohammadi (Intuit Engineering) talks about how crucial it is to guarantee the reliability and accuracy of ML models before deploying them to production. He shows how to streamline ML model deployment via automated sanity checks. A simple 4-step approach: Ensure online model scores sink to the output store, rescore offline, compare online to offline scores, & verify input data.
Expand your knowledge on the topic further with David Andrzejewski's talk where he explored how we can achieve reliable ML systems amidst their inherent complexities. A very insightful presentation that covers academic research, industry best practices, and software tools:
Catch your chance to share your thoughts! Submit your talk!