01 Batch PCA

"Centering" the dats for PCA is

• optional

• a must

• a nice to have

• What is centering?

• all negative

• all positive and real

• complex

• imaginary

The Principal Components (PCs) are invariant to scale differences between dimensions (T/F)?

• True

• False

If I scale the entire dataset X by a constant factor a=5, the directions of the PCs ...

• remain the same.

• change directions.

• increase length.

• decrease length.

What changes when I compute the PCs on a scaled version of X? (e.g. PCA(X) vs. PCA(5*X))

• Nothing.

• The Eigenvalue changes.

• The Eigenvectors change direction.

• The Eigenvectors change in magnitude.

PCA(X) vs. PCA(5*X): How will the Eigenvalue change?

Which is not a suitable application for PCA?

• dimensionality reduction

• data visualisation

• image classification

• compression

What do the PCs with highest λ represent?

• direction of smallest variance

• outliers in the data

• clusters in the data

• direction of highest variance

What are PCs with small variance good for?

• outlier detection

• measuring the noise component

• nothing whatsoever

• decorrelating the data

What is "whitening"?

• decorrelating the data

• sorting the eigenvalues in descending order

• normalizing the eigenvalues

• turning into a zombie with freekishly blue eyes