Opened 11 years ago

Closed 11 years ago

## #639 closed request (fixed)

# PCA on a Kernel Matrix

Reported by: | Peter | Owned by: | Peter |
---|---|---|---|

Priority: | major | Milestone: | yat 0.7 |

Component: | utility | Version: | trunk |

Keywords: | Cc: |

### Description

The PCA class works on a data matrix, i.e., it calculates the covariance matrix and diagonalizes it to find directions with most variance.

I need something slightly different because of two things. I wanna avoid holding the data matrix in memory because it is huge. Second, I have missing values in my data so I need to calculate the covariance matrix in a slightly different way than it is done in the PCA class (see AveragerPairWeighted).

I want to calculate the covariance matrix (or let's call it kernel matrix) outside the PCA class, feed it into a PCA class, calculate the principal components, and project my data onto say two largest principal components.

This doesn't really fit into the current PCA class, so I'm thinking of creating a new class PCA2 (?).

Thoughts?

**Note:**See TracTickets for help on using tickets.

(In [2324]) new class KernelPCA. closes #639