Fix handling of keys in epi_keys_mold

[mgyliu]

Creating this issue to track. I will implement this in my vignette PR (TODO mgyliu: link the PR)

Related to epiprocess issue cmu-delphi/epiprocess#182

Action items

• Change the way epi_keys_mold handles multiple other_keys in epi_df metadata

The problem

When the metadata for an epi_df looks like this

> attr(ex1,"metadata")
$geo_type
[1] "state"

$time_type
[1] "day"

$as_of
[1] "2020-06-03"

$other_keys
[1] "county_code" "another_key"

calling predict fails in epi_keys_mold because of the purrr::map_chr line. map_chr doesn't like that other_keys has a nested list in it.

Example to replicate

ex1_input <- tibble::tibble(
  geo_value = rep(c("ca", "fl", "pa"), each = 2),
  county_code = c("06059","06061","06067",
                  "12111","12113","12117"),
  another_key = 1:6, # <- I added this additional key 
  time_value = rep(seq(as.Date("2020-06-01"), as.Date("2020-06-03"),
                       by = "day"), length.out = length(geo_value)),
  value = 1:length(geo_value) + 0.01 * rnorm(length(geo_value))) %>% 
  tsibble::as_tsibble(
    index = time_value, 
    key = c(geo_value, county_code,another_key))

ex1 <- as_epi_df(x = ex1_input, geo_type = "state", time_type = "day", as_of = "2020-06-03")
attr(ex1,"metadata")

# Output 
# $geo_type
# [1] "state"
# 
# $time_type
# [1] "day"
# 
# $as_of
# [1] "2020-06-03"
# 
# $other_keys
# [1] "county_code" "another_key"

rec1 <- epi_recipe(ex1) %>%
  step_epi_ahead(value, ahead=3) %>%
  step_epi_naomit()

wf1 <- epi_workflow(rec1, parsnip::linear_reg()) %>% parsnip::fit(ex1)
stats::predict(wf1, ex1)

The last line spits out this error:

Error in `stop_bad_type()`:
! Result 3 must be a single string, not a character vector of length 2
Run `rlang::last_error()` to see where the error occurred.
Backtrace:
1: stop(fallback)
2: signal_abort(cnd, .file)
3: abort(message, x = x, expected = expected, actual = actual, what = what, 
4: stop_bad_type(x, expected, actual = actual, what = what, arg = arg, 
5: stop_bad_vector(x, expected_ptype, expected_length, what = what, 
6: purrr:::stop_bad_element_vector(c("county_code", "another_key"
7: purrr::map_chr(mold$extras$roles[molded_names %in% keys], names)
8: epi_keys_mold(mold)
9: grab_forged_keys(components$forged, components$mold, new_data)
10: predict.epi_workflow(wf1, ex1)
11: stats::predict(wf1, ex1)

The fix

The fix is to replace the last 2 lines with

mold_keys <- map(mold$extras$roles[molded_names %in% keys], names)
unlist(unname(mold_keys))

[dajmcdon]

Closed in #116