1
$\begingroup$

I was reading the principal component analysis section of ISLR. The authors define the $m$th principal component score from the $m$th principal loading vector $\phi_1=(\phi_{1m},...,\phi_{pm})$ as $Z_m=\phi_{1m}X_1+...+\phi_{pm}X_m$. On page 387-388 of the pdf, the author states that for an $n$ by $p$ (observations by predictor) design matrix $X$ that is centered, if $M = min(n-1,p)$ (number of principal components where the score has non-zero variance), then one can reconstruct the original design matrix $X$ using $x_{ij}=\sum_{m=1}^M z_{im}\phi_{jm}$ (where $1 \leq i \leq n$ and $1 \leq j \leq p$) which is similar to this answer.

I'm trying to get an intuition as to why we only need the principal components where the score has non-zero variance to reconstruct the original design matrix.

$\endgroup$

1 Answer 1

0
$\begingroup$

The principal component scores along any principal component loading vector average to 0 (because of centering - see page 384 of the pdf). If the variance is also 0, then all the scores must also be 0. Hence we can take only the principal components with nonzero variance score vector.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.