Happy Pi Day 2018!

In honor of Pi Day, I usually like to do a little on-topic code snippet.

This year I was running low on time, but I thought I’d ask the question “Can Pi be useful in predictive modeling, ML, AI, etc?”.

Of course the answer is going to be a big “yes!”. Transformations with natural numbers are underutilized, perhaps because it’s not always intuitive to leverage a constant scalar in a model. Let’s see a trivial example with the famous `iris` data set, built into R.

Compare the 2 models below and you’ll be pleasantly surprised. Pi helped us explain more variance and helped to create another highly significant predictor capturing a potentially unique effect:


summary(lm(I(iris$Species=="setosa") ~ iris$Sepal.Length, data = iris))
summary(lm(I(iris$Species=="setosa") ~ iris$Sepal.Length + I(iris$Sepal.Length^pi), data = iris))