I was recently working on some ML transformation code. The whole transformer/estimator thing that is pretty standard now in ML libraries. I got really frustrat

I was recently working on some ML transformation code. The whole transformer/estimator thing that is pretty standard now in ML libraries. I got really frustrated when I realized that a "pipeline" made of "transformers" is just a composed function where each individual function in the composition is curried/closed over its own parameters. I thought "why don't they just keep this simple and use functions instead of all these cumbersome classes!" Then I started to write something using function composition to keep things simple and got annoyed because sometimes I wanted some helper functions on my functions and it was cumbersome to do that. And I thought "I have made a mistake. The true nature of these things is objects." And then I didn't actually become enlightened. Because everything's an object which is also a closure and we're all made of stardust. But none of that means anyone is going to understand or use my library. Sometimes it's necessary to give things names people expect and keep the higher order abstractions to myself.