A recipe records ordered preparation separately from product identity, contracts, and destinations. It does not replace ordinary R transformation code. It makes a preparation sequence inspectable and reusable when that is valuable.
library(dataraft)
preparation <- dataraft.core::dr_recipe() |>
dataraft.core::dr_step_mutate(
premium = round(premium, 2)
) |>
dataraft.core::dr_step_filter(
status != "cancelled"
)
preparation<recipe> 2 preparation step(s)
# A tibble: 2 × 2
step operation
<chr> <chr>
1 mutate_1 dplyr::mutate
2 filter_2 dplyr::filterA recipe can be attached without executing it.
policies <- dr_product(
"policies",
read.csv("data/policies.csv")
) |>
dataraft.core::dr_add_recipe(preparation)
result <- dr_run(policies, write = FALSE)
dr_collect(result)# A tibble: 4 × 4
policy_id broker_id premium status
<int> <int> <dbl> <chr>
1 1 10 1200. active
2 2 20 950. paid_up
3 3 10 601. lapsed
4 4 30 1500 active The current recipe API includes familiar mutate, filter, select, rename, arrange, summarise, and distinct steps. dr_step_transform() provides an escape hatch for custom functions.
The book’s broker reference data is a meaningful dependency of the portfolio product.
brokers <- read.csv("data/brokers.csv")
portfolio_recipe <- dataraft.core::dr_recipe() |>
dataraft.core::dr_step_lookup(
brokers,
by = "broker_id",
name = "brokers"
)The output contract belongs to the data after preparation. If the lookup adds region, the published interface should say whether region is part of the product.
For a bounded set of mutate, select, and rename operations, DataRaft can inspect direct field dependencies without running the recipe.
lineage_recipe <- dataraft.core::dr_recipe() |>
dataraft.core::dr_step_mutate(
rounded_premium = round(premium, 0)
) |>
dataraft.core::dr_step_select(
policy_id,
rounded_premium
)
dataraft.core::dr_column_lineage(
lineage_recipe,
c("policy_id", "premium")
)$fields
$fields$policy_id
[1] "policy_id"
$fields$rounded_premium
[1] "premium"
$complete
[1] TRUE
$reason
NULLUnsupported transformations return complete = FALSE rather than a guessed map.
Use a recipe when the ordered preparation itself should be inspectable, composed, tested, or connected to lineage. Use an ordinary R function when a local transformation is clearer that way.
If a recipe grows into many transformations with several independently useful outputs, that is often a sign that the workflow contains more than one product.
Recipe expressions retain their R environments. Prefer explicit product sources for business data and stable constants for configuration. A recipe is inspectable preparation, not an isolation sandbox.