Cannot use poly(x) within a loop? 'degree' must be less than number of unique points

Question

library(ISLR)

Choosing a model based on any feature:

amodel <- lm(crim ~ poly(dat$zn, 3), data = Boston)

Works, returns a model. When I try to do this with every variable in a loop I get:

dat <- dplyr::select(Boston, -crim)
sapply(dat, function(x) {
  mod <- lm(crim ~ poly(x, 3), data = Boston)
  summary(mod)
})

Error in poly(x, 3) : 'degree' must be less than number of unique points

How can I get this to run within a loop? I would like to create a model for each poly(feature, 3) in Boston (minus crim feature).

score 1 · Accepted Answer · answered Jun 14 '18 at 06:27

1

Numerical overflow would be my guess. If you dig into the code for poly, you'll see that it generates the individual polynomial terms as an intermediate step:

X <- outer(x, seq_len(n) - 1, "^")

and when n (the degree of the polynomial you want) is 3, combined with the code used to generate the linear model lm(), the resulting terms go up to exponentially.

answered Jun 14 '18 at 06:27

dynamicwebpaige

373
1
10

So why does it work outside of a loop but not within one? The block for "amodel" above does run, just not within a loop – Doug Fir Jun 14 '18 at 06:48
1

@DougFir, it does not work for the `chas` column no matter within or out of the for loop. That's why you get the error. – mt1022 Jun 14 '18 at 07:12
d'oh! Thanks for spotting that – Doug Fir Jun 14 '18 at 07:18

Cannot use poly(x) within a loop? 'degree' must be less than number of unique points

1 Answers1

Linked