It's the line of best fit, not the line of good fit

From the best paper ever written by an undergrad

https://pages.cs.wisc.edu/~kovar/fittedHall.gif

One Line to rule them all
One Line to find them
One Line to bring them all
and in the data bind them

One line best-fits all

Machine Learning enthusiasts: Why settle for linear regression when you can deploy a Gradient-Boosted Random Deep Neural Net Surface Vector Cluster that consumes the entire power of Iceland to trace a perfect ∞-dimensional hypersphere around those blue points? Overparameterization is the future!

Fine! I'll use a second order polynomial to fit this instead. But that's the last order I'm willing to go to!

Oh, it's trending up. That's progress!

Always XKCD: https://xkcd.com/882/

R^2=0.03

Well, some students of mines actually put some figure like that in al lab report...

If you're having proportion of explained variance problems I feel bad for you son,
I got ninety nine problems but a fit ain't one.

(⌐■_■)

Dat spread tho

Gaussian: "Squint."

I don't know why you get credit at all; I do all the work. - Excel

Looks like a successfully trained learning model, to me. (Sarcasm)

They'll also try to linear fit even the most obvious exponential curve.

And if that doesn't work, there's always factor analysis.

How can you argue with a word like "best' anyway ;)

Probably minimal surface ellipse.