Keep original labels in `step_discretize`

[renanxcortes]

Hi there!

I'd like to ask for a feature that would keep the original labels generated by the internal cut function in discretize, instead of "bin1", bin2", etc. Perhaps adding an argument keep_cut_labels = TRUE, for example.

Minimal Reproducible Example:

Current Behaviour:

library(modeldata)
data(biomass)

biomass_tr <- biomass[biomass$dataset == "Training",]
biomass_te <- biomass[biomass$dataset == "Testing",]

rec <- recipe(HHV ~ carbon,
              data = biomass_tr) %>% 
  step_discretize(carbon)

rec <- prep(rec, biomass_tr)
binned_te <- bake(rec, biomass_te)
table(binned_te$carbon)

Expected behaviour:

breaks <- quantile(biomass_tr$carbon, probs = seq(0, 1, length = 4 + 1))
table(cut(biomass_te$carbon, breaks = breaks))

[juliasilge]

You can get those values out using the tidy() method for a recipe:

library(recipes)
#> Loading required package: 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
#> 
#> Attaching package: 'recipes'
#> The following object is masked from 'package:stats':
#> 
#>     step
data(biomass, package = "modeldata")

biomass_tr <- biomass[biomass$dataset == "Training",]
biomass_te <- biomass[biomass$dataset == "Testing",]

rec <- recipe(HHV ~ carbon,
              data = biomass_tr) %>% 
  step_discretize(carbon)

rec <- prep(rec, biomass_tr)
binned_te <- bake(rec, biomass_te)
table(binned_te$carbon)
#> 
#> bin_missing        bin1        bin2        bin3        bin4 
#>           0          22          17          25          16

tidy(rec, 1)
#> # A tibble: 5 x 3
#>   terms   value id              
#>   <chr>   <dbl> <chr>           
#> 1 carbon -Inf   discretize_gclTQ
#> 2 carbon   44.7 discretize_gclTQ
#> 3 carbon   47.1 discretize_gclTQ
#> 4 carbon   49.7 discretize_gclTQ
#> 5 carbon  Inf   discretize_gclTQ

^{Created on 2021-03-29 by the reprex package (v1.0.0)}

You can read more about tidying a recipe here. Can you say more about your use case for wanting factor levels like that in your output?

[renanxcortes]

Hi, Julia, thanks for the reply! I didn't know that you can tidy the recipe and get these values!

But to be more clear in my use case, it would be awesome if, after you bake a recipe with new data, you could keep the label of the "step_discretize" steps instead of "bin01", "bin02", "bin03", etc., with the labels being similar of the ones generated by the "cut" function.

This would facilitate the EDA exploration/understanding of the data after baking.

Is there a workaround for this?

Minimal reprex:

library(recipes)
data(biomass, package = "modeldata")

biomass_tr <- biomass[biomass$dataset == "Training",]
biomass_te <- biomass[biomass$dataset == "Testing",]

rec <- recipe(HHV ~ .,
              data = biomass_tr) %>% 
  step_discretize(all_numeric(), 
                  num_breaks = 10,
                  options = list(keep_na = T, na.rm = T)) %>% 
  step_mutate_at(all_numeric(), fn = as.factor) %>% 
  step_other(all_predictors()) %>% 
  step_unknown(all_predictors())

rec <- prep(rec, biomass_tr)
binned_te <- bake(rec, biomass_te) # Perhaps add here keep_original_label = T
head(binned_te)

Tibble generated:

[renanxcortes]

Also related with #157. If step_cut could allow the user to specify the number of bins instead of fixed values for cutting, I think this current issue would be solved.

[renanxcortes]

I realized that the behaviour of step_discretize_xgb (as well as step_discretize_cart) of the embed package (https://github.com/tidymodels/embed) return the desired labels of this issue.

Minimal reprex:

library(modeldata)
library(xgboost)
library(recipes)
data(biomass, package = "modeldata")

biomass_tr <- biomass[biomass$dataset == "Training",]
biomass_te <- biomass[biomass$dataset == "Testing",]

rec <- recipe(HHV ~ .,
              data = biomass_tr) %>% 
  step_discretize_xgb(all_numeric(), 
                      outcome = "HHV") %>% 
  step_mutate_at(all_numeric(), fn = as.factor) %>% 
  step_other(all_predictors()) %>% 
  step_unknown(all_predictors())

rec <- prep(rec, biomass_tr)
binned_te <- bake(rec, biomass_te)
head(binned_te)

Which returns:

The downside of this, is that it does not necessarily return bins with roughly uniform frequencies.

[renanxcortes]

Hello, folks, is there some ongoing effort of this improvement? Cheers!

[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.