null predictor space: "Error in zzs$coefficients[u, 1] : subscript out of bounds"

[adw96]

The cryptic error message Error in zzs$coefficients[u, 1] : subscript out of bounds arises when the single predictor variable is all zeroes. (This sounds trivial that may happen in automated large scale data analysis.)

### ideal behaviour
lm(rnorm(10) ~ rep(0, 10))
glm(rpois(10, 10) ~ rep(0, 10), family="poisson")
rigr::regress("rate",
              formula = rpois(10, 10) ~ rep(0, 10) + rnorm(10),
              data=df)
rigr::regress("mean",
              formula = rpois(10, 10) ~ rep(0, 10) + rnorm(10),
              data=df)

### error with unhelpful error message
rigr::regress("rate",
              formula = rpois(10, 10) ~ rep(0, 10),
              data=df)
rigr::regress("mean",
              formula = rpois(10, 10) ~ rep(0, 10),
              data=df)

[svteichman]

@adw96 what do you want to happen in this setting? Do you want to stop with an error about a predictor being all zeros, or to continue (therefore considering this current error a bug and fixing it)?

[adw96]

Hi from my phone, so sorry for being unhelpful… but can we mimic the behavior of the ideal functions I mentioned above? (I can’t see what they do rn, though, doh, sorry)

Lmk if that’s vague and I’ll investigate when I’m at my computer

[svteichman]

Oh yes, I didn't read the code snippet well enough. Will do!