Member-only story
Why A Flat Spot Isn’t Always The Bottom, A Friendly Hessian Guide
What A Pringle Can Teach You About The Hessian Matrix
The Hessian Matrix For People Who Never Liked Matrices
Put a marble on the exact middle of a Pringle and let go.
It stays. For one proud second the marble looks like it has found the bottom of the chip.
Then it rolls off the side.
Most of us left school with a simple rule in our heads. If the slope is zero, you’re at a top or a bottom. The Pringle breaks that rule, and the thing that explains why is a little box of four numbers called the Hessian.
I didn’t care about that box until it broke something of mine. I was fitting a small model that predicts yes or no answers, and by accident I gave it the same column of data twice. Python, the programming language I was using, didn’t crash. It didn’t even complain. It calmly reported that one of my results had an uncertainty of about 12 million, for a number that was roughly 0.5.
By the end you’ll know exactly why. You’ll also see that the marble, that 12 million, the “plus or minus 3 points” in polls and a famous line from a 1999 game are one idea.
