`maxnet()` sometimes fails with hinge features

[AMBarbosa]

With default parameters, maxnet() fails for some data frames:

dat <- read.csv("https://raw.githubusercontent.com/AMBarbosa/files/refs/heads/main/d.csv")
maxnet(p = dat[ , 1], data = dat[ , -1])
# Error in m[p == 1, ] : (subscript) logical subscript too long

I found that the error occurs within regfun(p, mm), at mm <- m[p == 1, ], because mm (the model matrix) has zero rows. The error does not occur if we do not use hinge features:

maxnet(p = dat[ , 1], data = dat[ , -1],
       f = maxnet.formula(p = dat[ , 1],
                          data = dat[ , -1], 
                          classes = "lqpt"))
# apparently OK

[AMBarbosa]

We found that the error occurs when a variable has no variation within the dataset that is passed to maxnet(). When models are computed within a workflow that passes a subset of the starting data (where all variables do have variation) to maxnet(), the user isn't easily aware of this. So, I think it would be very useful to have maxnet() exclude non-varying columns (maybe with a warning) before model.matrix().