Learn how to finish scoping an ML project by defining milestones and resources. You'll write specifications for ML metrics like accuracy and fairness, software metrics like latency and throughput, estimate business impact, and list what you need: data, personnel, integrations, labeling timelines. Build a proof-of-concept first if specs are hard to estimate.
Say you've identified a problem, found a worthy solution, finished diligence for technical feasibility and value, and you think this is worth doing? Let's take a look at the last steps of the scoping process of Milestones and resourcing Determining milestones and resourcing involves writing out the key specifications for your project. This will include machine learning metrics such as accuracy or precision recall. For some applications this may also include fairness types of metrics. The specification will often also include software metrics regarding the software system, such as latency, throughput, queries per second, and so on, given the computational resources available. And if possible, you might also write out estimates of the business metrics you hope to move for the project you're scoping, such as how much incremental revenue if you have a way of estimating that. In addition, writing out the resources needed, how much data from which teams, personnel, any help you need from cross functional teams, what software integrations, data labeling support or other support you need from other teams, and finally the timeline on which you hope to achieve certain milestones or deliverables. Now, if you're looking at a machine learning project and you find that you're having a very hard time writing out some of these key specifications, then you might also consider carrying out a benchmarking exercise to compare it to other similar projects that others may have worked on before, or building a proof of concept first in order to get a better sense of what accuracy, precision or latency, throughput or other metrics might be feasible. And only after you've done that PLC proof of concept to then use that information to more confidently scope out the milestones and resources needed for a larger scale execution of the project you have in mind. So that's it. Congratulations on making it to the end of this section on scoping. I hope that these ideas will help you to pick valuable and meaningful projects to work on. And thank you also for sticking with me from today deployment, through modeling, through training, all the way back to scoping. And I hope this framework of the full cycle of machine learning project will also be useful for all the projects I hope you will build and deploy.
