synth_spec: Create a 'synth_spec' object

In tidysynthesis: A Common API for Synthesizing Data

Create a synth_spec object

Description

The synth_spec object holds specifications for modeling and sampling components for sequential synthetic data generation. Each component has an associated ⁠construct_*⁠ function called when creating a presynth object.

Usage

synth_spec(
  default_regression_model = NULL,
  default_classification_model = NULL,
  custom_models = NULL,
  default_regression_steps = NULL,
  default_classification_steps = NULL,
  custom_steps = NULL,
  default_regression_sampler = NULL,
  default_classification_sampler = NULL,
  custom_samplers = NULL,
  default_regression_noise = NULL,
  default_classification_noise = NULL,
  custom_noise = NULL,
  default_regression_tuner = NULL,
  default_classification_tuner = NULL,
  custom_tuners = NULL,
  default_extractor = NULL,
  custom_extractors = NULL,
  invert_transformations = TRUE,
  enforce_na = TRUE
)

Arguments

default_regression_model	A model_spec object from library(parsnip) for use in regression models.
default_classification_model	A model_spec object from library(parsnip) for use in classification models.
custom_models	A list of named lists each with two elements: vars for variable names, and model for their associated model. from library(parsnip).
default_regression_steps	A list of recipe::step_ function(s) from library(recipes) for use in regression models.
default_classification_steps	A list of recipe::step_ function(s) from library(recipes) for use in classification models.
custom_steps	A list of named lists each with two elements: vars for variable names, and steps for their associated recipe.
default_regression_sampler	A sampling function for drawing new values from regression models.
default_classification_sampler	A sampling function for drawing new values from classification models.
custom_samplers	A list of named lists each with two elements: vars for variable names, and sampler for their associated sampler
default_regression_noise	A noise function for adding noise to numeric values.
default_classification_noise	A noise function for adding noise to classification values.
custom_noise	A list of named lists each with two elements: vars for variable names, and noise for their associated noise
default_regression_tuner	A tuner from library(tune) for use in regression models.
default_classification_tuner	A tuner from library(tune) for use in classification models.
custom_tuners	A list of named lists each with two elements: vars for variable names, and tuner for their associated tuner
default_extractor	An optional method for extracting workflows or extracts from workflows.
custom_extractors	A list of named lists each with two elements: vars for variable names, and extractor for their associated extractor
invert_transformations	A Boolean for if outcome variable transformations applied through recipes should be inverted during synthesis. recipes need ids that begin with "outcome".
enforce_na	A Boolean for if NA values should be added into the synthetic data with enforce_na() during synthesis. An alternative approach is to add the NA values after synthesis

Value

A synth_spec object

Examples

rpart_mod <- parsnip::decision_tree() |>
  parsnip::set_engine(engine = "rpart") |>
  parsnip::set_mode(mode = "regression")

lm_mod <- parsnip::linear_reg() |> 
  parsnip::set_engine("lm") |>
  parsnip::set_mode(mode = "regression")

step1 <- function(x) {
 x |>
   recipes::step_center(recipes::all_predictors(), id = "center")
}

step2 <- function(x) {
  x |>
    recipes::step_scale(recipes::all_predictors(), id = "scale")
}

step3 <- function(x) { x |> step1() |> step2() }


synth_spec(
 default_regression_model = rpart_mod,
 custom_models = list(
   list("vars" = c("var1", "var2"), 
        "model" = lm_mod)
 ),
 default_regression_steps = step1,
 custom_steps = list(
   list("vars" = c("var2", "var3"),
        "steps" = step2),
   list("vars" = c("var4"), 
        "steps" = step3)
 ),
 default_regression_sampler = sample_rpart,
 custom_samplers = list(
   list("vars" = c("var1", "var2"), 
        "sampler" = sample_lm)
 )
)

tidysynthesis documentation built on March 17, 2026, 1:06 a.m.