kerasnip_step_view: View a Single Forecast Step of a Multistep kerasnip Fit

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

View a Single Forecast Step of a Multistep kerasnip Fit

Description

A kerasnip multistep (vector-valued) regression model returns a nested .pred list-column: one inner tibble per row, with a .step column plus one prediction column per forecasted variable. tailor/probably expect a single flat numeric .pred column instead — tailor::check_variable_type() requires is.numeric() on the outcome/estimate columns, which a list-column fails outright.

kerasnip_step_view() wraps a fitted multistep workflow together with one forecast step (and, if more than one variable is forecast, which variable), presenting it as an ordinary single-output fit: predict() returns a flat .pred column for that step alone.

Usage

kerasnip_step_view(x, step, var = NULL)

Arguments

x	A fitted (trained) workflow whose model is a multistep regression model (see create_keras_sequential_spec()/ create_keras_functional_spec() with a vector-valued output).
step	An integer, the forecast step to view.
var	A string, the forecasted variable to view. Required only if the model forecasts more than one variable; inferred otherwise.

Details

Unlike kerasnip_output_view(), a multistep model's per-step outcome columns (e.g. lead_1_value) are recipe-engineered from a single raw column via step_lead() — they are not present in a user's raw data the way genuine multi-output columns are. kerasnip_step_truth() recovers the true future value at a given step by re-baking the fitted recipe on raw data, which is what int_conformal_split() uses internally for this class.

probably::int_conformal_full() is also supported (see int_conformal_full.kerasnip_step_view()), with a materially different design from kerasnip_output_view()'s: refitting for a candidate value at this step means substituting it into the single raw column step_lead() derives every step's truth from, which shifts every nearby row's target too. It is only supported when step_lead() and step_sequence() share a single source column, matching all of kerasnip's own multistep examples.

Value

A kerasnip_step_view object.

Examples

## Not run: 
fit_obj <- fit(wf, data = train_data) # a multistep forecasting workflow
step_2 <- kerasnip_step_view(fit_obj, step = 2)
predict(step_2, new_data = test_data) # -> a single `.pred` column

## End(Not run)

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