I regularly deal with data in which I have a single metric that is computationally expensive to calculate. I also have numerous (less than a dozen) low-resolution metrics that attempt to approximate different elements of the expensive metric. I'd like to take a relatively small training set for which I have both the expensive and inexpensive metrics (say, 1000 or so instances) and come up with a model that predicts the high-resolution data from the low. All of my metrics are real numbers on a continuous scale. It is likely that some of the inexpensive metrics will be correlated.

I'm completely ignorant to virtually all machine learning techniques, but assumed this would be a common enough scenario that there would be some "canned" way of doing this (my naive hope was a simple online tool would exist). In the absence of that, what would be a straightforward way of accomplishing my goal?

  • 4
    $\begingroup$ Have you tried linear regression? Predicting a continuous quantity doesn't get much more "canned" than that. $\endgroup$ – Sycorax May 29 '15 at 20:17
  • $\begingroup$ Yeah.... Before wanting the most sofisticated, hard, and ussualy brute-force methods, try the linear ones. Even more, if you go onto more advanced methods, having not a good put knowledge wont put you on a very good position to interpret them... $\endgroup$ – Brethlosze May 30 '15 at 0:38
  • $\begingroup$ Start with a simple ARX method.... that is the most simple step to give... $\endgroup$ – Brethlosze May 30 '15 at 0:39

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.