This post is a tongue-and-cheek response to Professor writes a love song to statistics.
I decided to ask a non-linear model to rebut this musical march against violations of additivity and homogeneity of scaling. What more timely non-linear model to ask than ChatGPT?
I asked ChatGPT:
Write me a song about how general linear models are not the best. Please end it with “It is easy to see that this song was crap, but I’d like to see a GLM do better than that.”
It responded by naming the chat session “Why GLM is Inferior” and generated these lyrics.
Verse 1:
The general linear model, it’s a common sight
But it’s not the answer to all of our data’s plight
It’s got limitations that are plain to see
So it’s time to take a look, at what’s meant to be
Chorus:
The GLM is not the best, it’s time to confess
There’s more to data analysis than regression tests
From non-linear relationships to overfitting woes
Better models are available, so let’s take a look at those
Verse 2:
When it comes to complex data, the GLM’s in strife
It just can’t handle the curves in our life
And when it comes to outliers, it just falls apart
So it’s time to choose a model that’s got heart
Chorus:
The GLM is not the best, it’s time to confess
There’s more to data analysis than regression tests
From non-linear relationships to overfitting woes
Better models are available, so let’s take a look at those
Bridge:
From decision trees to random forests, we can do more
And neural nets with deep learning, will always score
So don’t be afraid to try a new approach
Better models are waiting, it’s time to encroach
Chorus:
The GLM is not the best, it’s time to confess
There’s more to data analysis than regression tests
From non-linear relationships to overfitting woes
Better models are available, so let’s take a look at those
Outro:
It is easy to see that this song was crap,
But I’d like to see a GLM do better than that.
So let’s not be afraid to try a new approach
Better models are out there, so let’s encroach!
Well, there you have it. ChatGPT can generate song lyrics from a short prompt. It kind of accommodated the request for a specific sentence at the end, although it wasn’t quite at the end but rather in the outro. I’d say it kind of missed the boat with “overfitting woes”. While GLMs can have overfitting problems, jumping over to neural networks or other general function approximators is not going to improve overfitting. If anything it will make things worse. But overall I am impressed with this output.