Lecture 10
Video
PCA
SVD
- produces same new axes as PCA
- when I try to reduce some dimensions, I loose some information
-
we need to minimize that loss, that’s what we do by removing least variance data
- PCA is unsupervised
- PCA is not good for labelled data - class labels wale, bcz overlap aa jayega
- best fit ke perpendicular line
- so 2nd best
