ability to extract the fitted polynomial/spline
aanilpala opened this issue · comments
I was reading the paper which the implementation of this library is based on. Quoting from it:
In this paper we extend lattice regression, which is a spline method with fixed knots on a regular grid and a linear kernel (Garcia et al., 2012), to be monotonic
My question is whether it is possible to obtain the polynomial explicitly in order to use in a gradient-based optimization procedure. I can't seem to find any such interface that allows this to happen.
Background: I have a number of calibrated lattice regressors and I need to jointly optimize them for the variable which the models are monotonic for.
On a second thought, I see that the polynomial being used depends on the input point and its distance to the surrounding lattice points so a fixed estimator formula like in a typical parametric model can't be obtained. I would be glad if someone can confirm this though.