One of the problems with being an analyst is you get used to traveling at 30,000 feet. You take briefings with vendors and speak with customers in the rarefied air of marketing PowerPoints and strategic plans without every spending time in the trenches. Time on the ground is important. If you do not actually spend time getting your hands dirty you become complicit in the thinking that abstraction is the great simplifier, and we replace understanding with tech vendor mythology. Every analyst is guilty of this at some point in their career, and this is especially true when it comes to AI.
So I decided to get dirty and try some of this for myself. Where to begin? Well certainly it would make sense to see what Microsoft, Google and Amazon have to offer. Amazon has a very academic approach, and they offer a number of lecture and demo style courses on ML designed to introduce you to their products. Google offers a “Crash Course on ML that also looked interesting, so that goes on the list as well. I am a big fan of hands on learning, so a host of web tutorials get added to the mix. While I worked as a developer for many years, most of my real commercial experience was with COBAL, C++, and SQL. That meant I would need a quick tune-up to build my Python skills. Luckily, Python is a joy to code in, and Pycharm, and Visual Studio make learning the syntax a breeze. Microsoft offers a great set of tutorials on Python coding, and after a few hours I felt like my coding mojo was coming back.
Now to get familiar with the tools – so many tools! Anaconda, TensorFlow, NumPy, Pandas, SciPY, scikit-learn, and matplotlib all get added to the environment. This took a good half day to figure out and configure, and when my first actual linear regression ran at the end of day one I felt like I was winning! I know, every noob goes through this, but I really felt like I was finally getting beyond the vendor hype and beginning to gain a deeper level of understanding. There is no better way to learn than to do. While these are the first baby step for any data scientist, I felt like I had made a big leap into the reality of producing actionable insights. Why do this? For as long as I have been involved in industrial software we have talked about analytics as a necessary tool for the road ahead. Databases, data warehouses, data marts, data lakes have always been better when we could get usable output to guide our decision making. Unfortunately, the promise almost always exceeded reality when this technology was passed into the customer’s hands. For practitioners in the supply chain space, the latest iteration of this hype cycle is the use of IoT and digital twins in the supply chain. Machine learning is how we unlock this value – and I am determined to figure this out. I will update you on my progress as I move forward – and I promise we will get back to supply chain in my next post 😊.