If features are x, y, z, then this transformer only generates (with order=3) columns for x, y, z, xy, yz, xz, xyz, ie, no feature appears with degree higher than one, and there is no constant term. I suppose that is what "interaction" means?
But what about the common use case of polynomial regression on a single feature? You want 1, x, x^2, x^3, which this transformer won't give.
Perhaps we can add a parameter interaction=true which when false adds all the monomials up to specified order.
cc @olivierlabayle
If features are
x,y,z, then this transformer only generates (withorder=3) columns forx,y,z,xy,yz, xz, xyz, ie, no feature appears with degree higher than one, and there is no constant term. I suppose that is what "interaction" means?But what about the common use case of polynomial regression on a single feature? You want
1,x,x^2,x^3, which this transformer won't give.Perhaps we can add a parameter
interaction=truewhich whenfalseadds all the monomials up to specified order.cc @olivierlabayle