A timely, or rather timeless, post! We are bombarded with classes and webinars to “learn Python in x days|weeks”. I recently completed a 7 week “Introduction to machine learning with Python” course at a local university and my takeaways dovetail with the points in your article. I’ve also checked out how other schools, online webinars, and meetup groups handle this same material and they all seem to be missing the process awareness and improvement that I took away from your comments on Steve McConnell’s book.

We all have a tendency to try to hurry through the “learning” stuff so we can get to the “doing” stuff, but the two are really complementary components of the same process. The course I took was great for getting exposure to how Python is being used to solve machine learning problems, but the coding we did was frankly more like kids playing with a shiny new gadget and less like the fault-tolerant, design-for-enhancement-or-replacement development that we need as ML becomes ubiquitous in our homes, workplaces and places in between!

I read “Code Complete” as a young developer in the early 90’s (ugh, dating myself a bit), and I thought it ironic that such brilliant software engineers came out of a company which at the time had a rep for spaghetti code in core products like windows. As I gained experience I realized that even with really smart people, the software development lifecycle is hard to manage well.

ML can eat up cpu and memory, even more of a reason for new practitioners to be made aware of good software process. I’ve heard the argument that due to Moore’s Law it doesn’t pay to spend time on code efficiency, but I’ve never believed that quality is something you can just turn on or off like a light switch. I agree, we need a more mature, software engineering approach with Python and ML!