augment with do fails on lm(y ~ poly(x))

[fja062]

I am getting an error when I use augment within do. My quantile regression model uses splines::bs(...) as the predictor, but the problem can be replicated with lm using poly.

library(MASS)
data(mcycle)

bind_rows(a = mcycle, b = mcycle, .id = "ver") %>%
   group_by(ver) %>% 
   do({
     fit = lm(accel ~ poly(times, degree = 2), data = .)
     augment(fit)
   })

Error in bind_rows_(x, .id) : Argument 2 must be length 133, not 266

[dgrtwo]

The problem is that in the augment output, the poly(times, degree = 2) column isn't a vector, but a 2-column matrix.

This is definitely a problem for augment in general (should we always force the output columns from model.frame() to be vectors? What kind of error could we give?) But it has a workaround in this case of giving the data argument to augment(), which means you get the original columns like accel and times and not a poly() matrix column:

bind_rows(a = mcycle, b = mcycle, .id = "ver") %>%
   group_by(ver) %>% 
   do({
     fit = lm(accel ~ poly(times, degree = 2), data = .)
     augment(fit, data = .)
   })

[fja062]

Ah ok, that makes sense. And the workaround works perfectly - thanks!

[alexpghayes]

My take is that augment should add columns to original data but not keep any columns from model.frame. I think this specification resolves this issue, although we'll need to check that it gets enforced.

[alexpghayes]

Closing in favor of #405. Eventually matrix-column support in tibbles will arrive and this will just work.

[jcdoesds]

I am having difficulty seeing how to apply this solution to nested data. In that case, If I declare the augment with data = . or data = .x, I get an error.

test <- mtcars %>%
  nest(data = -cyl) %>%
  mutate(
    fit = map(data, ~ lm(mpg ~ disp + poly(wt, 2), data = .x)),
    tidied = map(fit, tidy),
    glanced = map(fit, glance),
    augmented = map(fit, augment) 
    # augmented = map(augment(fit, data = .)) # fails w subscript out of bounds
  )
test %>%
  unnest(augmented) # this fails

test$augmented[[1]]# this works

After some discussion on SO, I see that we can get the output via:

test %>%
   unnest_wider(augmented)

[github-actions]

This issue has been automatically locked. If you believe you have found a related problem, please file a new issue (with a reprex: https://reprex.tidyverse.org) and link to this issue.