ML-Ops (Machine Learning Operations)
A machine learning project that doesn’t make it to production is an expensive failure, yet this is the reality for many. By treating each “phase” of work - business objective, modelling, scoring - as separate from each other, the risk of each transitioning smoothly to the next phase becomes more challenging.
The software development industry suffered analogous problems for decades. Dev-Ops thinking, processes and tools have moved the considerations of evaluation, deployment, monitoring and maintenance upfront, as first-class citizens of the software development life cycle. Data science programmes often stumble for the same reasons that software projects used to: too little emphasis on coding at the expense of productionising the solution.
Failed AI projects may have suffered the same waterfall approaches that the software development world has replaced with agile, meaning that most of the work has taken place before the project is aborted. This is the most expensive way to fail. Agile practices work with shifting requirements, experimentation, iterative delivery and, significantly, landing product in the hands of customers early and often.
ML-Ops promises to bring stability and efficiency to machine learning to see the time from conception to deployed solution decimated.
Robotica Machine Learning brings the best of Free, Open-Source Software (FOSS) Dev-Ops practices to the machine learning arena, utilising Gherkin, Jenkins and Git to automate and streamline the evolution of models over time.
Request a demo