ML Canvas - Requirements Template for Machine Learning Projects

Before you start your work on the model, you should ask questions - to the stakeholders, and to yourself. These questions will help in having a clear vision and understanding of why the model is needed, what are the technical requirements, and how you will be making one.

When you're making a decision with a model, if you have the answers to these questions on back of your head, chances are that you won't make wrong ones.

I have seen Data Scientists create a model having 90-95% accuracy but are so complex that those couldn't be deployed either because they were now no longer interpretable or these models were using more resources than initially planned that it is no longer viable for the company to deploy a model. This usually results in waste of time and redoing of a lot of work.

So what are these questions that you should ask?

These questions are divided into business problem questions and technical questions. Business problem questions include questions like "what is the use case?" or "why is it needed?" etc. and technical questions include "which metric to optimise?" or "would deployment of this model require change in company's tech infrastructure?" or "how much it is going to cost to keep model in production?" etc.

To navigate all these questions, Louis Dorard came up with a Machine Learning Canvas that breaks down all the questions into different categories. I have made an excel template of the same that you can use for your Machine Learning projects in the future.

To download, click on "I want this!" button.