Skip to content

Recipes Internals

Source: vignettes/articles/internals.Rmd
internals.Rmd

These elements should not be accessed by users directly. They should instead be accessed via helper functions.

The recipe object will contain different elements depending on whether it has been prep()ed or not. We will list all of them here and what they contain.

An unprepped recipe will contain:

  • var_info
  • term_info
  • steps
  • template
  • levels
  • retained
  • requirements
  • ptype
  • strings_as_factors

A prepped recipe will contain:

  • var_info
  • term_info
  • steps
  • template
  • retained
  • requirements
  • ptype
  • strings_as_factors
  • tr_info
  • orig_lvls
  • fit_times
  • last_term_info

var_info

Appears in both unprepped and prepped recipes

A tibble containing information about the original data set columns. It will contain the 4 columns variable, type, role, and source.

  • variable is a character vector of variable names.
  • type is a list column of character variables. These character vectors are used in type selection functions such as all_numeric() and all_integer(). Each variable has one or more character values to allow for granular selections. An integer variable will have type values c("integer", "numeric") allowing it to be selected by all_numeric() and all_integer() but not all_doubles().
  • role is a character vector of roles.
  • source is a character vector. It can contain 2 values "original" and "derived". "derived" is used to denote variables that are created by steps such as those created by step_dummy() and step_pca().

Note that there can be duplicates of variable as each variable can have multiple roles as denoted by role.

The source element for unprepped recipes should be "original" for all variables. This is kept here for parity with term_info.

term_info

Appears in both unprepped and prepped recipes

A tibble that contains the current set of terms in the data set. This initially defaults to the same data contained in var_info. This is updated after each step is prep()ed. It should contain the state of the variables you would see after bake()ing a recipe.

The format is the same as described in var_info.

steps

Appears in both unprepped and prepped recipes

A list of step or check objects that define the sequence of preprocessing operations that will be applied to data. The default value is NULL.

The content of each element will depend on each step or check object.

template

Appears in both unprepped and prepped recipes

A tibble of the data is passed to recipe(). This is initialized to be the same as the data given in the data argument but can be different after the recipe is trained. If retain = TRUE is set when prep()ing then the template dataset will contain the data after it has been processed by the recipe. retain = TRUE is the default.

levels

May appear in both unprepped and prepped recipes

When a recipe is unprepped this element will be NULL. When it has been prepped it will either not be present or not. If at least one of the variables in the template has factor levels. Otherwise, it will be missing, not NULL. If the element is there it will be a named list, with the names matching those in template. Each element will contain the elements values and ordered, with some also including factor. The factor element will only appear for variables that are factors, unordered or ordered.

  • values A character vector of factor levels if the variable is a factor, NA otherwise.
  • ordered A logical value, TRUE if the variable is an ordered factor, FALSE if the variable is an unordered factor, NA otherwise.
  • factor A logical value, appears to always be TRUE.

retained

Appears in both unprepped and prepped recipes

A logical value. Will be NA for unprepped recipes. Otherwise will depend on what value was passed to retain in prep().

requirements

Appears in both unprepped and prepped recipes

A named list. This will be an empty list unless update_role_requirements() is used. If update_role_requirements() has been used it will gain an element named bake() which will contain a named logical vector. The names correspond to the roles that have updated requirements, and the value itself is whether the role is required at bake() time.

ptype

Appears in both unprepped and prepped recipes

A zero-row slice of the data set is passed to recipe(). This data set is used with recipes_ptype() and recipes_ptype_validate() to make sure the recipe takes on the same shape of data when specified as when prepped and baked.

strings_as_factors

Appears in both unprepped and prepped recipes

A single logical. Whether strings should be converted to factors in prep(). Defaults to TRUE.

tr_info

Appears only in prepped recipes

A data.frame with 1 row and 2 variables nrows and ncomplete.

  • nrows is an integer, the number of rows in the training data set used in prep().
  • ncomplete is an integer, the number of rows that don’t contain any missing values in the data that was passed to prep().

Note that ncomplete represents the counts before any of the steps are applied. tr_info is used for the print method for the recipe itself.

orig_lvls

Appears only in prepped recipes

This element is highly related to levels and will thus have the same structure. levels contains the information regarding the final data set. orig_lvls contains the same structure of information, but regarding the data set it was trained on before the steps take effect. This information is in many ways similar to what is found in ptype.

fit_times

Appears only in prepped recipes

A tibble containing two columns stage_id and elapsed. It will contain 2 rows for every step or check contained in the recipe.

  • stage_id A character vector on the format {stage}.{name}_{id} where {stage} can be either prep or bake, name will be the name of the step and id its id element.
  • elapsed A double, a calculation done with proc.time() to estimate how long each step took to apply.

This element is used by extract_fit_time().

last_term_info

Appears only in prepped recipes

A grouped tibble, this will look similar to var_info and term_info. The main difference is that there is one row per variable in the data. Variables with multiple roles don’t have different rows and instead have them all listed in the list element of `role.

  • variable is a character vector of variable names.
  • type is a list column of character vectors.
  • role is a list column of character vectors.
  • source is a character vector. It can contain 2 values "original" and "derived".
  • number is an integer corresponding to the last step where that variable was available.
  • skip is a logical corresponding to the value of skip in the last step where that variable was available.

This data set is used In case a variable was removed, and that removal step used skip = TRUE, we need to retain its record so that selectors can be properly used with bake. This is important for skipped steps which might have resulted in columns not being added/removed in the test set.