#swiss-roll

Question 1: Single choice

mediumAI-checkedDimensionality Reduction

Your 3-D data lie on a Swiss roll: a 2-D sheet rolled up into a spiral, with no direction of the space that is useless. Why is PCA down to 2 dimensions a poor choice here?

It can only project onto a plane, which squashes the roll instead of unrolling it1
Two principal components always keep less than half of the variance of 3-D data2
PCA needs labels to find directions that separate the different layers of the roll3
PCA cannot be fitted on data with fewer features than samples4

Timed test