pull_workflow_fit() in R: Extract a Model From a Workflow

The pull_workflow_fit() function in R returns the fitted parsnip model_fit object stored inside a trained workflow. It is the older name for extract_fit_parsnip() and still works, with a deprecation warning.

pull_workflow_fit(wf_fit)              # extract the parsnip model_fit
extract_fit_parsnip(wf_fit)            # modern, non-deprecated replacement
tidy(pull_workflow_fit(wf_fit))        # coefficient table from the fit
pull_workflow_fit(wf_fit)$fit          # reach the raw engine object
pull_workflow_spec(wf_fit)             # extract the model spec instead
pull_workflow_preprocessor(wf_fit)     # extract the recipe or formula

What pull_workflow_fit() does

pull_workflow_fit() unwraps a trained workflow and returns the parsnip model inside it. A workflow bundles a preprocessor (a formula or recipe) and a model specification. When you call fit() on it, the fitted model is stored inside the workflow object. pull_workflow_fit() reaches in and hands you that fitted model.

The object you get back is a parsnip model_fit, the same class you would get from calling fit() on a bare parsnip specification. That means every parsnip tool works on it: predict(), tidy(), glance(), and augment().

The function comes from the [workflows package](https://workflows.tidymodels.org/reference/extract-workflow.html), not parsnip. It was renamed to extract_fit_parsnip() in workflows 0.2.3 to match a family of extract_*() generics shared across tidymodels. The old name still works but prints a deprecation warning on every call.

pull_workflow_fit() syntax and arguments

The signature takes a single argument because the function only locates an object. It does no computation; it walks into the fitted workflow and returns one element.

The return value is a parsnip model_fit. Its class also carries the engine, so a workflow fit with the lm engine returns an object of class c("_lm", "model_fit").

Passing an unfitted workflow raises an error. There is no model to extract until fit() has run, so always call pull_workflow_fit() on the output of fit(), never on the workflow() you assembled.

Extract a workflow model: four examples

Each example uses the built-in mtcars dataset so you can run it directly. Start by assembling and fitting a linear regression workflow.

Example 1 extracts the parsnip fit. The returned object carries both the engine tag and the model_fit class.

Example 2 produces a tidy coefficient table. Once you hold the model_fit, parsnip and broom tools apply directly.

Example 3 reaches the raw engine object. The model_fit stores the native model under $fit, so you can drop to base R tools when needed.

Example 4 switches to the current name. Replace pull_workflow_fit() with extract_fit_parsnip() and the warning disappears.

pull_workflow_fit() vs extract_fit_parsnip()

The two functions are interchangeable; only the name and the warning differ. extract_fit_parsnip() belongs to a consistent extract_*() family that also covers recipes, engines, and model specifications.

Use extract_fit_parsnip() in all new code. Reach for extract_fit_engine() instead when you need the native object directly, for example to call summary() or confint(), which dispatch on the engine class rather than on model_fit.

Common pitfalls

Three mistakes account for most pull_workflow_fit() errors.

• Calling it on an unfitted workflow. pull_workflow_fit(car_wf) fails because car_wf was never trained. Pass the result of fit(), not the assembled workflow.
• Expecting the raw engine object. pull_workflow_fit() returns a parsnip model_fit, not an lm. Functions like confint() need the native object, reached with $fit or extract_fit_engine().
• Ignoring the deprecation warning. The warning is harmless today, but a future workflows release may remove the function. Migrate to extract_fit_parsnip() now.

Try it yourself

Related workflows functions

The pull_workflow_*() and extract_*() families cover every part of a workflow.

• extract_fit_parsnip(): the current name for pull_workflow_fit().
• extract_fit_engine(): returns the raw engine model instead of the parsnip wrapper.
• extract_recipe(): returns the prepped recipe used by the workflow.
• extract_spec_parsnip(): returns the model specification held by the workflow.
• extract_preprocessor(): returns the formula or recipe before prepping.

FAQ

Is pull_workflow_fit() the same as extract_fit_parsnip()?

Yes. extract_fit_parsnip() is the renamed version of pull_workflow_fit(), introduced in workflows 0.2.3. Both functions take a fitted workflow and return the identical parsnip model_fit object. The only practical difference is that pull_workflow_fit() prints a deprecation warning. Any code using the old name keeps working, but the tidymodels team recommends the extract_*() family for all new projects.

Why does pull_workflow_fit() show a deprecation warning?

The workflows package renamed its pull_workflow_*() accessors to extract_*() so they would match a shared set of generics across tidymodels packages. The old functions were kept for backward compatibility but marked as deprecated. The warning is a reminder to migrate; it does not stop your code from running.

How do I get the raw lm or ranger object from a workflow?

pull_workflow_fit() returns a parsnip model_fit wrapper, not the native model. To reach the underlying object, either index into the $fit element of the result or call extract_fit_engine() on the workflow. The engine object carries its own class, so base R tools such as summary() and plot() then dispatch correctly.

Can I call pull_workflow_fit() on a workflow before fitting it?

No. The function only locates an already-trained model inside the workflow. An assembled but unfitted workflow has no model to return, so the call raises an error. Always run fit(workflow, data) first and pass that result to pull_workflow_fit() or extract_fit_parsnip().