Principal Components

1. Given data matrix $X$ with $N$ rows (samples) and $k$ columns (features) – assume each feature has mean zero.
2. The matrix $Q=\frac{1}{N}X^{\intercal}X$ is symmetric and its entries are the variances/covariances.
3. If $v$ is a vector, then $Xv$ is called a “score” – a synthetic measure of the data.
4. The variance of the score is $v^{\intercal}Qv$.
5. Critical points of variance are eigenvectors of $Q$.
6. These critical directions are called “principal components”.

---