Avoiding the tiresome training & test data split

modelling
Published

September 3, 2018

I really don’t like splitting data into ‘train’ and ‘test’. I don’t mean that I’m against the idea of it, though you could say it’s a waste of data that could be used to better your model, but I mean that actual assignment in R of ‘train’ and ‘test’. I always liked destructuring in Python, and I like it a lot in 2018 JavaScript, so when I remembered that the zeallot package has it, I thought it would be a good opportunity to see how that could fit in a tidy modelling pipeline. Much to my delight, one little helper function later, it works perfectly.

Loading packages and data:

library(recipes); library(rsample); library(tidyverse); library(zeallot)

# load some data:
data("credit_data")

Next comes our little helper function:

m_bake <- function(recipe_object, data){
  cd <- initial_split(credit_data)
  tr <- training(cd)
  te <- testing(cd)
  x1 <- bake(recipe_object, tr)
  x2 <- bake(recipe_object, te)
  return(list(x1, x2))
}

Now you get a nice clean pipeline with a train/test split as a result, using zeallot’s great %->% operator:

recipe(Status ~ ., data = credit_data) %>%
  step_knnimpute(all_predictors()) %>%
  step_center(all_numeric()) %>%
  step_dummy(-all_numeric()) %>%
  prep() %>%
  m_bake(data = credit_data) %->% c(train, test)

ls()
## [1] "credit_data" "m_bake"      "test"        "train"
head(train); head(test)
## # A tibble: 6 x 23
##   Seniority  Time    Age Expenses Income Assets  Debt Amount Price
##       <dbl> <dbl>  <dbl>    <dbl>  <dbl>  <dbl> <dbl>  <dbl> <dbl>
## 1      1.01  13.6  -7.08    17.4   -12.6 -5429. -343. -239.  -617.
## 2      9.01  13.6  20.9     -7.57  -10.6 -5429. -343.  -38.9  195.
## 3     -7.99  13.6 -13.1      7.43   40.4 -2929. -343. -139.  -138.
## 4     -7.99 -10.4 -11.1     -9.57  -34.6 -5429. -343. -729.  -553.
## 5     -6.99  13.6  -1.08    19.4    72.4 -1929. -343. -389.   182.
## 6     21.0   13.6   6.92    19.4   -16.6  4571. -343.  561.   337.
## # … with 14 more variables: Home_other <dbl>, Home_owner <dbl>,
## #   Home_parents <dbl>, Home_priv <dbl>, Home_rent <dbl>,
## #   Marital_married <dbl>, Marital_separated <dbl>, Marital_single <dbl>,
## #   Marital_widow <dbl>, Records_yes <dbl>, Job_freelance <dbl>,
## #   Job_others <dbl>, Job_partime <dbl>, Status_good <dbl>
## # A tibble: 6 x 23
##   Seniority   Time      Age Expenses Income Assets   Debt Amount  Price
##       <dbl>  <dbl>    <dbl>    <dbl>  <dbl>  <dbl>  <dbl>  <dbl>  <dbl>
## 1      2.01 -10.4    8.92      34.4   58.4  -2429.  -343.   961.  1522.
## 2     -1.99   1.56  -3.08       4.43 -16.6  -1429.  -343.   111.   114.
## 3     25.0  -22.4   30.9        9.43  58.4   -429.  1657.  -439.  -113.
## 4     -6.99  13.6   -6.08     -20.6   -4.65 -5429.  -343.   211.  -210.
## 5     -7.99   1.56  -0.0804   -20.6  -23.8  -4629.  -343.   461.   387.
## 6      6.01 -22.4   13.9       19.4   56.4  -4429.  -343.  -589.  -813.
## # … with 14 more variables: Home_other <dbl>, Home_owner <dbl>,
## #   Home_parents <dbl>, Home_priv <dbl>, Home_rent <dbl>,
## #   Marital_married <dbl>, Marital_separated <dbl>, Marital_single <dbl>,
## #   Marital_widow <dbl>, Records_yes <dbl>, Job_freelance <dbl>,
## #   Job_others <dbl>, Job_partime <dbl>, Status_good <dbl>

Lovely.

(Ok, I still had to split it (!!), but once you write this function, you can just call this.)