Is there an online filter algorithm that generates an optimal set (maximizes correlation of prediction) of feature weights (ie relative feature scaling) for a simple 1NN (or KNN) regression using either Manhattan or Euclidean distance metrics? I have a set of numerical X and y values, and I'd like to create a simple nearest neighbor predictor, but I'm unsure how to optimize the feature weighting. I tried r-relieff, but it seems like brute force forward feature selection works better (but is too slow). Is there a way I can reformulate into a qp problem when both X and y are all numerical and can be solved efficiently using an online scheme (ie single pass similar to r-relieff)?


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.