Modeltime workflowsets fit

allows working with workflow sets and modeltime. Combination of recipes and models are trained and evaluation metrics are returned.

modeltime_wfs_fit(.wfsets, .split_prop, .serie)

Arguments

.wfsets	workflow_set object, generated with the workflow_set() function from the 'workflowsets' package.
.split_prop	time series split proportion.
.serie	time series dataframe.

Value

tbl_df containing the model id (based on workflow_set), model description and metrics on the time series testing dataframe. Also, a .fit_model column is included, which contains each fitted model.

Details

Given a workflow_set containing multiple time series recipes and models, adjusts all the possible combinations on a time series. It uses a split proportion in order to train on a time series partition and evaluate metrics on the testing partition.

See also

sknifedatar website

Examples

library(dplyr)
library(earth)

data <- sknifedatar::data_avellaneda %>% mutate(date=as.Date(date)) %>% filter(date<'2012-06-01')

recipe_date <- recipes::recipe(value ~ ., data = data) %>% 
  recipes::step_date(date, features = c('dow','doy','week','month','year')) 

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

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

sknifedatar::modeltime_wfs_fit(.wfsets = wfsets, 
                               .split_prop = 0.8, 
                               .serie = data)
#> ── MODEL: R_date_M_mars 
#> ✓ Training finished OK.
#> 
#> 
#> ── 1 models fitted ♥ ───────────────────────────────────────────────────────────
#> 
#> ── 0 models deleted x ──
#> 
#> # A tibble: 1 x 10
#>   .model_id  .model_desc .type   mae  mape  mase smape   rmse   rsq .fit_model  
#>   <chr>      <chr>       <chr> <dbl> <dbl> <dbl> <dbl>  <dbl> <dbl> <list>      
#> 1 R_date_M_… EARTH       Test  7577.  6.79 0.542  6.33 12524. 0.553 <model_time…