Saturday, December 04, 2010
PCA for optimization?
To do a multiobjective optimization over chi-square statistics of simulated data versus experimental data, I wonder if it would make sense to weight the individual objectives using the eigenvalues (standardized or not) of the covariance matrix of the different stats? That is, use PCA to assign weights in a non-arbitrary way. Hmm...
Subscribe to:
Post Comments (Atom)
This to me seems logical, provided that your objectives do not "overlap." Else you would have two of your objectives crammed into the same PC. Of course, you could then apply the weights of the objectives to that PC to extract them, I suppose.
ReplyDeleteI am eager to see your data and play visually with it. It's hard for me to picture "protein network space" yet but if I could just find a way to think of it geostatistically that would be fun.