DataTau logo


new | ask | show | submit
Machine Learning in Production. Where 99% of companies screw up (
7 points by thegurus 219 days ago | web | 4 comments

Launching in production is a product life cycle more than a data science project. Key success factors include availability of labeled data (willingness to label), ability to make process changes that flow from ML predictions, and clear business value linked to appropriate stakeholders. Each is a non-trivial task that can constitute 90% of the overall data science product development effort.


Some great points made. In my experience, screwups tend to start when - the "prototype" / minimum viable product can't scale. N^2 algorithms have a cost -- and it can be high! - the data source is not-repeatable/updating/lacking lineage

Yeah, so many models made from a data source that is a .csv that someone updates manually twice a year, no way to put that to production