Closed RaulRomaniF closed 5 years ago
This package only implements (bi)linear models. On some datasets, a rank 2 fit with a GLRM can provide an excellent visualization; on others, it doesn't. If you want something that's designed for visualization, try t-SNE?
With nonlinear GLRM I mean to use a multilayer perceptron to solve for X and Y in an alternating minimization fashion. With visualization I mean dimensionality reduction to 2D.
I think a nonlinear GLRM would improve the reconstruction, but do you think it will also improve the low dimensional representations "X variable"?.
What kind of model (i.e. losses and regularizations) do you think it would be good for visualization.