Notes on building machine learning systems

digging for requirements

  • what exactly is the business objective?
    • how does the company expect to use and benefit from this model?

workflow to Approach a ML problem -> Prototype

  • what kind of question or goal we wanna answer

  • how to define and measure success -> like using a business metric like increased profit or decreased losses

  • acquire the data and build a working prototype - a loop [TODO]

    • analyze the mistakes
    • collect more or diff data
    • change the task formulation slightly
  • humans in the loop

    • algotithms might increase response time or reduce cost
    • TODO

From Prototype to Production

  • data analytics teams
  • production teams -> reimplement the solution for robust, scalable system
    • offline evaluation
    • online testing using A/B testing

reference