-
Make multidimensional categorical hinge loss
-
Set up a single command to: a) pick m random subsets of data, b) choose model of rank k, regularization constant α for each subset, c) impute missing data from each of the m selected models.
- (nandana will do this)
-
Call julia from R
- i write a wrapper
- nandana writes the UX
-
Documentation!
- how to think about mpca
- imputation
- error metrics for cross validation
- new syntax for fitting data frame
- how to specify loss function(s)
- parallel fitting
- full rank model / prisma
-
Poisson loss
- scaling?
- to log or not to log? that is the interpretative issue