Neural Networks in Awk

There has been tremendous push for Python, R, Scala, and Matlab in Data Science. And while I definitely see values in all these tools, I specifically feel as if there is nothing truly groundbreaking going on here. I was very happy to see this book on shelves, that appears to feels the same way I do. The sub-title is what stuck out to me: facing the future with time-test tools.

Why do you need new(er) tools such as Python to do nothing more than classical mathematical problem solving? Really, you don’t - there’s no magic sauce in these languages that edges out computational ability of traditional tools such as those found in Unix. If nothing else, you’re getting a strong community, rich libraries, and being on the bleeding edge of what’s new in data science; and that’s worth a lot, but you’re also paying tremendous performance and portability costs, which people like myself pay attention to.

I did some quick googling to find out if anyone had ever wrote a simple neural network in Awk (circa 1977, but still widely used today), unsurprisingly I didn’t find anything. I’m sharing my code here for anyone who may have some interest in doing a neural network in a fast interpretive language, with a much smaller tool chain than that of modern interpretive languages.

Awk takes on several different software names… awk, gawk, nawk, mawk. The syntax across them all is the same - just the underlying code of the interpreter has been changed over the years. Whatever you choose to use is entirely up to you.

If I get some extra time I’ll eventually benchmark some comparisons and post them here.