Unsupervised Learning
Dimensionality Reduction
9 min+80 XPStep 1/11
PCA: keep the directions that carry the signal
PCA finds the direction of maximum variance in your data — the axis along which points spread out most — then the next axis perpendicular to it, and so on.
Project onto just the top few of those axes and you keep the bulk of the structure in a fraction of the dimensions. It's linear, fast, the axes come ranked by importance — and it powers visualization, denoising, and speed-ups for every downstream model.
insight
Most real datasets aren't actually as high-dimensional as they look — the points live near a lower-dimensional surface. PCA's job is finding that surface.