recipes step_date() in R: Extract Date Features for Modeling

The recipes step_date() function in R turns a Date column into calendar features such as year, month, and day-of-week so a model can learn from them. You add it to a recipe(), then apply it with prep() and bake().

step_date(rec, order_date)                              # default: dow, month, year
step_date(rec, order_date, features = c("year"))        # year only
step_date(rec, order_date, features = c("doy","week"))  # day-of-year and week
step_date(rec, order_date, label = FALSE)               # numeric month and dow
step_date(rec, order_date, abbr = FALSE)                # full names like Monday
step_date(rec, all_date_predictors())                   # every date column
prep(rec) |> bake(new_data = NULL)                       # learn, then apply

What step_date() does in R

step_date() expands a single Date column into several predictor columns. A raw date carries information a model cannot read directly: seasonality lives in the month, weekly patterns live in the day of week, and long-term trend lives in the year. step_date() pulls those parts into their own columns so a regression or tree can use them.

By default it extracts three features, dow, month, and year, and names the new columns <variable>_<feature>. A column called order_date becomes order_date_dow, order_date_month, and order_date_year. The day-of-week and month columns come back as factors with readable labels, while the year is an integer.

step_date() syntax and arguments

step_date() attaches a date-expansion operation to a recipe. You pass the recipe first, then the date columns to expand, chosen by name or with the all_date_predictors() selector.

The arguments you will actually touch:

The features argument accepts year, doy (day of year), week, decimal (decimal date), semester, quarter, dow, month, and mday (day of month).

Extracting date features: worked examples

Build the recipe, prep it, then bake. A recipe is a plan until prep() registers the steps and bake() applies them. This example uses the default three features on the order_date column.

Three new columns appear and order_date is kept, because keep_original_cols defaults to TRUE. To pull different parts, pass a features vector. Here we ask for day of year, week number, quarter, and the decimal date.

By default dow and month arrive as factors. Many engines, especially glmnet and xgboost, want numeric input, so set label = FALSE to get integers instead.

step_date() vs lubridate vs base R

Pick the tool that matches where the work happens. All three extract calendar parts, but only step_date() keeps the logic inside a reusable recipe.

Use step_date() when the feature extraction belongs to a modeling pipeline, because the recipe travels with the workflow and applies the exact same transformation to test and production data. Reach for lubridate inside a one-off mutate() during exploration, and use base format() only when you cannot add a dependency.

Common pitfalls with step_date()

Most failures trace back to column type or to the leftover Date column. Watch these three traps when adding the step.

1. Passing a character column. step_date() needs class Date or POSIXct. A column of strings like "2023-01-15" triggers an error. Convert it first with as.Date(), or add step_mutate() ahead of step_date().
2. Feeding the Date column to the model. Because keep_original_cols defaults to TRUE, order_date survives baking. Most engines reject the Date class, so add step_rm(order_date) after step_date() once the features exist.
3. Expecting numbers but getting factors. With the default label = TRUE, dow and month are factors. A linear or boosted model needs numeric input, so either set label = FALSE or follow with step_dummy().

Try it yourself

Related recipes steps

step_date() is one of several recipes steps for time-based predictors. These pair naturally with it in a tidymodels workflow:

• step_holiday() flags whether each date is a named holiday.
• step_lag() creates lagged copies of a column for time-series features.
• step_window() computes rolling statistics over a moving window.
• step_dummy() converts the new month and dow factors into 0/1 columns.
• step_rm() removes the original Date column once the features exist.

FAQ

What does step_date() do in R?

step_date() is a recipes step that expands a Date column into separate calendar features. By default it extracts the day of week, month, and year, creating columns named <variable>_dow, <variable>_month, and <variable>_year. You add it to a recipe(), register it with prep(), and apply it with bake(). Because calendar parts are deterministic, the step needs no training data to compute its output.

What features can step_date() extract?

The features argument accepts any of year, doy (day of year), week, decimal (decimal date), semester, quarter, dow (day of week), month, and mday (day of month). The default is c("dow", "month", "year"). Pass a character vector to choose others, for example features = c("quarter", "week"). Each requested feature becomes one new column appended to the data.

Does step_date() keep the original date column?

Yes. The keep_original_cols argument defaults to TRUE, so the source Date column remains in the baked output alongside the new feature columns. This is convenient for inspection but causes errors when the data reaches a model, because most engines cannot handle the Date class. Add step_rm() after step_date() to drop the original column before fitting.

What is the difference between step_date() and step_holiday()?

step_date() extracts general calendar parts such as month and day of week from a date. step_holiday() answers a narrower question: it creates 0/1 indicator columns marking whether each date falls on a named holiday, like LaborDay or NewYearsDay. They are complementary, so a typical recipe runs step_date() first for seasonality, then step_holiday() for special-day effects, then step_rm() to discard the raw date.