Leaning Mathematical Function

GhostriderGhostrider Member Posts: 60 Contributor II
edited November 2018 in Help
Say I have a set of values that describe the function f(x,y)=z.  What would be the best learning algorithm for predicting other values of the function values (not necessarily learning the function's expression)?  I guess basically this is a multi-variable regression problem.


  • Options
    steffensteffen Member Posts: 347 Maven
    You identified the problem correctly as regression problem. Furthermore there is no "best" algorithm for such a generic task, due to the no free lunch theorem. Any learner which can handle numerical labels could be "the one".

    Have fun trying them out.
  • Options
    rakirkrakirk Member Posts: 29 Contributor II
    Steffen is right, but I had an idea that may be helpful- Paramater optimization will probably help a lot too.

    My thought was that the regression may be optimal when consdering a more localized view of he data rather than considering all training data at once- maybe it will more quickly approach a good estimate of the real value this way. Try looking at various portions of the data for the regresson, ie. 'window' the data into different sized test sets. This may be a bit like creating a movig average or differeniation in learning. What are everyone's thoughts on this?

Sign In or Register to comment.