DUPLEX with external test set

[zecojls]

Hi Leo,

Is it possible to extend the DUPLEX algorithm with external test sets? The idea is to sample a primary bigger SSL given a small secondary SSL with both sharing the underlying spectral properties. The required parameters would be the X primary, X secondary, and the subset (n) size.

Cheers

[zecojls]

I've been playing with this idea and found a way to search for similar samples. Do you think that this makes sense?

1. Run PCA on primary and get scores (P).
2. Project secondary onto primary space and get scores (S).
3. Calculate covariance (C) of the first p principal components of the secondary (I set p=20).
4. Estimate Mahalanobis distance of P given covariance C from S:
   mahalanobis(P, center = FALSE, C)
5. Filter top n samples given the Mahalanobis distance.

PCA of secondary (red) projected onto primary space (black):

After running the algorithm, neighbor samples (n=15000) are selected: