My dependent variable is a multi- (say 10-20) dimensional vector; the obvious measure of distance between two points is L2-norm and I'd be happy to use that. My explanatory variables (features) are either categorical or continuous. Is there a regression tree function/class that would allow for multi-dimensional y? fitrtree unfortunately requires a scalar.

0 |

12/21/2016 4:30:03 PM

On 12/21/2016 11:30 AM, Pawel wrote: > My dependent variable is a multi- (say 10-20) dimensional vector; the > obvious measure of distance between two points is L2-norm and I'd be > happy to use that. My explanatory variables (features) are either > categorical or continuous. Is there a regression tree function/class > that would allow for multi-dimensional y? fitrtree unfortunately > requires a scalar. Can you simply fit a decision tree to each dimension in your dependent vector separately? You mention distance between two points but I am not sure what you mean by that distance nor how the choice of L2 is relevant. Regression tree minimizes mean squared error for scalar response, if that helps. -Ilya

0 |

12/21/2016 8:48:47 PM