Get the best workflow for each time series — modeltime_wfs

obtains the best workflow for each time series based on a performance metric.

modeltime_wfs_multibestmodel(.table, .metric = NULL, .minimize = TRUE)

Arguments

.table	a tibble that comes from the output of the `modeltime_wfs_multifit()` or `modeltime_wfs_multiforecast()` functions. For the `modeltime_wfs_multifit()` function, the 'table_time' object must be selected from the output.
.metric	a string of evaluation metric, the following symmetrical can be supplied: 'mae', 'mape','mase', 'smape','rmse','rsq'.
.minimize	boolean (default = TRUE), TRUE if the error metric should be minimized, FALSE in order to maximize it.

Value

a tibble, corresponds to the same tibble supplied in the '.table' parameter but with the selection of the best workflow for each series.

Examples

library(dplyr)
library(earth)

df <- sknifedatar::emae_series

datex <- '2020-02-01'
df_emae <- df %>%
  dplyr::filter(date <= datex) %>% 
  tidyr::nest(nested_column=-sector) %>% 
  head(2)

receta_base <- recipes::recipe(value ~ ., data = df %>% select(-sector))

mars <- parsnip::mars(mode = 'regression') %>% parsnip::set_engine('earth')

wfsets <- workflowsets::workflow_set(
  preproc = list(
    R_date = receta_base),
  models  = list(M_mars = mars),
  cross   = TRUE)

wfsets_fit <- modeltime_wfs_multifit(.wfs = wfsets,
                                     .prop = 0.8, 
                                     serie = df_emae)
#> ✓ Workflow training finished OK.
#> ✓ Workflow training finished OK.
#> 
#> ── 1 models fitted ♥ ───────────────────────────────────────────────────────────
#> 
#> ── 0 models deleted x ──
#> 

sknifedatar::modeltime_wfs_multibestmodel(.table = wfsets_fit$table_time,
                                          .metric = "rmse",
                                          .minimize = TRUE)
#> # A tibble: 2 x 6
#>   sector  nested_column   R_date_M_mars nested_model    calibration   best_model
#>   <chr>   <list>          <list>        <list>          <list>        <list>    
#> 1 Comerc… <tibble [194 ×… <workflow>    <model_time [1… <model_time … <int [1]> 
#> 2 Ensena… <tibble [194 ×… <workflow>    <model_time [1… <model_time … <int [1]>