predict.kerasnip_model_fit: Predict Method for kerasnip Model Fits

In kerasnip: A Bridge Between 'keras' and 'tidymodels'

Predict Method for kerasnip Model Fits

Description

S3 method for predict() dispatched on kerasnip_model_fit objects. Before delegating to the standard parsnip predict machinery, it checks whether the underlying model type is registered in the current parsnip session. If not (e.g. after loading a saved workflow in a new R session), it transparently replays the full parsnip registration using metadata stored on the spec object — requiring no manual step from the user.

Usage

## S3 method for class 'kerasnip_model_fit'
predict(object, new_data, ...)

Arguments

object	A kerasnip_model_fit object.
new_data	A data frame of predictors.
...	Passed to the parsnip predict method.

Details

The metadata needed for re-registration (kerasnip_layer_blocks, kerasnip_functional) is embedded on the spec object by the spec constructor function at call time. This means it is preserved across saveRDS()/readRDS() and bundle()/unbundle() round-trips.

For full model weight portability (i.e. to be able to predict() on new data in a new R session), use bundle::bundle() before saving. Plain saveRDS() preserves the spec structure and will auto-register, but the underlying Keras model weights are not portable without bundling.

Value

A tibble of predictions.

kerasnip documentation built on Sept. 4, 2026, 1:06 a.m.