I have a covariance matrix of a standardized data set. Doing a singular value decomposition i find near zero singular values and would therefore like to truncate it.
I know of Picard plots which would do the trick. But I have only used it on systems such as $\textbf{d}=\textbf{Gm}$ when doing least squares inversions.
Does anyone know a good technique I could use to determine the truncation level of a decomposed covariance matrix?
Common rule of thumb is truncating the analysis whenever the ratio of the singular values exceeds $0.9$, i.e., stop when you first have $\sum_{i=1}^k \sigma_i / \sum_{i=1}^p, \sigma_i \ge 0.9$.