Yesterday I attended for the first time the Machine Learning Fireside Chats meetup run by the BBC Blue Room. Even though it was quite far from work and home I could not help going after seeing the name of the speakers on the line up.

Simon Raper (Founder of Coppelia Machine Learning and Analytics), Giles Pavey (Head of Data Strategy at the Department for Work and Pensions) and Dr. Shahzia Holtom (Lead Data Scientist at Pivotal Labs) got together yesterday to discuss about what it is important when getting started with a Machine Learning project.

The panelist discussion evolved around the significance of following points at the beginning of a project and the importance we should be giving to each of them.

  • Expectations
  • Budget
  • Readiness of the data
  • Skills/People
  • Tools
  • Complexity/Simplicity of algorithms

Expectations. There are two sides on this. Some companies do not think they are ready or they do not want to apply machine learning but there is always a way to demonstrate the value to them. And once they have seen it, they believe it is magic.

Others, think machine learning would solve all the problems, and since it lies under the AI umbrella their expectations are too high.

It is important to demystify machine learning, explain what it is and what it can and can not do.

Budget. Obviously, a budget is needed for any machine learning project, but there is no need of a huge investment in order to show and prove the impacto of data science projects.

Readiness of the data. It is important to have some data to work with, but ,with the necessary skills, you should be able to get enough data to get started almost in any situation. It can be old data, data on silos or weirdly formatted data, whatever it is, you do not need a perfect infrastructure that provides you a lot of Big Data for your model. We are talking about getting started, and a lot can be done with a “small” sample of the data.

Skill/People. This is by far the most stressed aspect. The need of good people, yes with good skill, but mainly good people. Attitude and problem solving skills are mentioned as the most important factors for a person to be a successful data scientist. Technical skill are useful but they can also be learn and you can always train the right people.

Tools/Machine learning as a service. There is no need of fancy or expensive tools to get started. You can do machine learning in open source programs. But, what about the machine learning as a service tools? Well, there are ups and down with it. Even if you have a machine learning as a service software you will still need to decide what problem and how do you want to solve it. You need to know which is the best machine learning approach for it and how should you treat your data for it. It is a great opportunity to do machine learning projects faster and more efficiently but not yet a standalone tool that an work without and expert managing it.

Complexity/Simplicity of algorithms. Start always small. The key thing is to show the value first, usually a simple regression is enough to show an approximation of the output and how it can impact the business. It is also good for the stakeholders to understand what it does at first and then built upon and continue improving it.

Summing up, the right people is all you need to get started. A good, motivated and problem solver Data Scientist can provide value using machine learning on a project.

I found this session really enjoyable and meaningful and I will definetly be attending the next Machine Learning Fireside Chats!

Why not join us?