The gist of my message is: don’t focus on language, technique, methods, or even the methodology. Focus on having a broad exposure to various types of problems, languages, techniques and the resourcefulness to find a good approach when needed.

In the tech-world they use the phrase “technology stack” to refer to the various technologies that go into the development of a full solution. Similarly, a good predictive modeling solution may involve connecting the dots between different types of algorithms, visuals, possibly even programming languages. Investing too deeply into one layer of this predictive modeling stack will leave you with less time to explore the rapidly expanding landscape of algorithms, methods, and languages – and fewer options when it comes time to designing your solution. Fewer options = higher chances of a sub-optimal solution.

Don’t worry about not going deep or not knowing the mathematical underpinnings of a given method. Being resourceful is more important. Depth is easily accessible through search and the wonderful efforts of modelers worldwide that publish their work. What is not easy is knowing what to search for in the first place. Obviously you would need a certain level of mathematical know-how in order to absorb the information that you search for, however you don’t need to carry around the detailed specifics of a given algorithm in your head.

Know a little bit about a lot of things, and know how to access deeper knowledge about the one thing when you apply it in your predictive modeling solution.

The post Learning Machine appeared first on Predictive Modeler.