Generating from bootstraps causing .rep clash

[robjhyndman]

library(fable)
library(feasts)
library(dplyr)

data <- USAccDeaths %>%
  as_tsibble()
stl <- data %>%
  model(stl = STL(value))
# Bootstrap variants of the data
sim <- stl %>% 
  generate(new_data = data, times=3) %>%
  select(-.model)
# Fit models to each bootstrap variant
simulated_fits <- sim %>%
  model(ets = ETS(.sim)) 
# Generate future sample paths from each model
simulated_fits %>% 
  generate(h=10, times=20)
#> New names:
#> * .rep -> .rep...1
#> * .rep -> .rep...4
#> Error: Can't subset columns that don't exist.
#> x Columns `.rep` and `.rep` don't exist.

^{Created on 2020-12-18 by the reprex package (v0.3.0)}

Because .rep is part of the key structure, just renaming it also causes problems.

[mitchelloharawild]

The key structure needs to update for the conflict resolving column names done by {vctrs} - not sure why this isn't happening yet.

However in the short term, renaming should work without issue:

library(fable)
#> Loading required package: fabletools
library(feasts)
library(dplyr)
#> 
#> Attaching package: 'dplyr'
#> The following objects are masked from 'package:stats':
#> 
#>     filter, lag
#> The following objects are masked from 'package:base':
#> 
#>     intersect, setdiff, setequal, union

data <- USAccDeaths %>%
  as_tsibble()
stl <- data %>%
  model(stl = STL(value))
# Bootstrap variants of the data
sim <- stl %>% 
  generate(new_data = data, times=3)
# Fit models to each bootstrap variant
simulated_fits <- sim %>%
  # This can be changed at the tsibble level, or later...
  # rename(variant = .rep) %>% 
  model(ets = ETS(.sim)) 
# Generate future sample paths from each model
simulated_fits %>% 
  # Additionally we can retain the model identifier for the bootstrap if we desire
  rename(variant = .rep, bootstrap_model = .model) %>%
  generate(h=10, times=20)
#> # A tsibble: 600 x 6 [1M]
#> # Key:       bootstrap_model, variant, .model, .rep [60]
#>    bootstrap_model variant .model    index .rep    .sim
#>    <chr>           <chr>   <chr>     <mth> <chr>  <dbl>
#>  1 stl             1       ets    1979 Jan 1      8085.
#>  2 stl             1       ets    1979 Feb 1      7408.
#>  3 stl             1       ets    1979 Mar 1      8398.
#>  4 stl             1       ets    1979 Apr 1      8600.
#>  5 stl             1       ets    1979 May 1      8978.
#>  6 stl             1       ets    1979 Jun 1     10061.
#>  7 stl             1       ets    1979 Jul 1     10491.
#>  8 stl             1       ets    1979 Aug 1     10067.
#>  9 stl             1       ets    1979 Sep 1      8826.
#> 10 stl             1       ets    1979 Oct 1      8953.
#> # … with 590 more rows

^{Created on 2021-01-07 by the reprex package (v0.3.0)}

[mitchelloharawild]

Related MRE:

library(fable)
#> Loading required package: fabletools
library(dplyr)
#> 
#> Attaching package: 'dplyr'
#> The following objects are masked from 'package:stats':
#> 
#>     filter, lag
#> The following objects are masked from 'package:base':
#> 
#>     intersect, setdiff, setequal, union
mdl <- as_tsibble(cbind(mdeaths, fdeaths)) %>% 
  model(SNAIVE(value))
bind_cols(mdl, as_tibble(mdl)["key"])
#> New names:
#> * key -> key...1
#> * key -> key...3
#> Error: Can't subset columns that don't exist.
#> x Column `key` doesn't exist.

^{Created on 2021-01-07 by the reprex package (v0.3.0)}

[mitchelloharawild]

Consolidating with #275