Linear Regression Uses All Four Subspaces
An interactive live film: in the space of targets, y splits into a prediction in C(X) and a residual in N(X^T); in the space of coefficients, a redundant feature makes N(X) a line along which predictions do not change. Find the coefficients with no null-space part.
Narration
Three samples and three features. The third feature is the first one measured in other units: exactly twice it.
On the left is the space of targets. The columns of X span the blue plane, the column space, and the target y sits off it.
The rest of the narration is in the book's Companion.
