▶Variance as signal
A useful projection preserves the long direction of a correlated cloud.
Principal Component Analysis: Motivation
Finding the direction that carries the signal
Many datasets contain several measured variables but far fewer independent patterns. PCA rotates the coordinate system so that the strongest pattern becomes visible first.
watchone idea
→manipulateone example
→leave withone intuition
The essential idea: dimension reduction works when the data vary much more along some directions than others.
One Concept · One Example
What to notice
- PCA looks for directions, not original variables.
- The first component captures the largest variance.
- Projection discards low-variance directions.
Rotate first, then compress
Reconstruction
Keeping only the leading direction gives the closest rank-one approximation in squared distance.
Which direction preserves the most information?
Rotate the projection axis through a correlated cloud. Watch captured variance rise as reconstruction error falls.
PCA finds a compact coordinate system by preserving the directions of greatest variation.