---
title: "Example scripts: training and frozen workflows"
description: "Training, own fit/predict functions, fixed engine options, encoding, reference splits, independent Inductive sizes, full-model transfer, pairwise OvO and frozen adoption."
output:
  rmarkdown::html_vignette:
    toc: true
    toc_depth: 2
vignette: >
  %\VignetteIndexEntry{Example scripts: training and frozen workflows}
  %\VignetteEngine{knitr::rmarkdown}
  %\VignetteEncoding{UTF-8}
---

Train complete tidymodels workflows, preserve names with explicit schemas, or
calibrate already fitted workflows. Install the optional workflow dependencies
listed in `DESCRIPTION`; individual examples also identify their engine and
assessment dependencies.

Start with `vignette("workflows", package = "vennabers")` for the
step-by-step guide. `?vennabers_examples` indexes all four example collections.

## Choose an example

- [Training binary and shared multiclass workflows](#calibrated-workflow)
- [Encoding binary names](#encoded-binary)
- [Encoding multiclass and regression names](#encoded-multiclass-regression)
- [Reference-style data roles](#reference-splits)
- [Independent Inductive training and calibration sizes](#inductive-sizes)
- [Own fit and predict functions](#model-adapter)
- [Fixed engine options and internal validation](#engine-options)
- [Binary full-model transfer](#full-model-workflow)
- [Shared multiclass full-model transfer](#full-multiclass-workflow)
- [Regression full-model transfer](#full-regression-workflow)
- [Pairwise Inductive OvO training](#pairwise-inductive)
- [Pairwise Cross OvO training](#pairwise-cross)
- [Calibrating a frozen binary workflow](#frozen-binary)
- [Calibrating frozen multiclass and regression workflows](#frozen-multiclass-regression)
- [Keeping the original schema for frozen adoption](#frozen-schema)

## Running an example

Each heading below names an installed script. To inspect and run one:

```{r run-example, eval=FALSE, purl=FALSE}
script <- system.file("examples", "calibrated-workflow.R", package = "vennabers")
file.show(script)
source(script)
```

Replace the filename to run another script. The listings below show the complete
code without executing it during documentation builds. Sourcing a script runs
its fits and checks; resampling, tuning and composition matrices can take longer
than the introductory guides.

## Training binary and shared multiclass workflows {#calibrated-workflow}

`calibrated-workflow.R`: Retain local preprocessing, models and calibration together, then inspect predictions and learning diagnostics.

```{r calibrated-workflow, eval=FALSE, purl=FALSE, code=readLines(system.file("examples", "calibrated-workflow.R", package = "vennabers"))}
```

## Encoding binary names {#encoded-binary}

`encoded-binary.R`: Preserve reserved predictor and class names through encoding, native resampling and decoded predictions.

```{r encoded-binary, eval=FALSE, purl=FALSE, code=readLines(system.file("examples", "encoded-binary.R", package = "vennabers"))}
```

## Encoding multiclass and regression names {#encoded-multiclass-regression}

`encoded-multiclass-regression.R`: Use explicit schemas for shared OvO/OvR and regression with Inductive/Cross calibration and native evaluation.

```{r encoded-multiclass-regression, eval=FALSE, purl=FALSE, code=readLines(system.file("examples", "encoded-multiclass-regression.R", package = "vennabers"))}
```

## Reference-style data roles {#reference-splits}

`reference-splits.R`: Illustrate ordered holdouts and explicit fold complements on synthetic data, without claiming paper benchmark reproduction.

```{r reference-splits, eval=FALSE, purl=FALSE, code=readLines(system.file("examples", "reference-splits.R", package = "vennabers"))}
```

## Independent Inductive training and calibration sizes {#inductive-sizes}

`inductive-sizes.R`: Request 50 training and 20 calibration rows out of 100, leaving 30 unassigned rows that only a full-model refit learns from, then resolve the same sizes on every class pair of a pairwise multiclass fit.

```{r inductive-sizes, eval=FALSE, purl=FALSE, code=readLines(system.file("examples", "inductive-sizes.R", package = "vennabers"))}
```

## Own fit and predict functions {#model-adapter}

`model-adapter.R`: Wrap GLM, multinomial and linear regression fit/predict functions with `model_adapter()`, calibrate them like native engines and keep their settings bound for saved fits. The GLM callback also carries pairwise OvO, Sigmoid postprocessing and a grid search over a recipe degree and the VA loss.

```{r model-adapter, eval=FALSE, purl=FALSE, code=readLines(system.file("examples", "model-adapter.R", package = "vennabers"))}
```

## Fixed engine options and internal validation {#engine-options}

`engine-options.R`: Set fixed rpart options with `parsnip::set_engine()`, check that the full-model refit carries them, see extra fit arguments rejected, and read the not-run XGBoost early-stopping call.

```{r engine-options, eval=FALSE, purl=FALSE, code=readLines(system.file("examples", "engine-options.R", package = "vennabers"))}
```

## Binary full-model transfer {#full-model-workflow}

`full-model-workflow.R`: Compare matched and full prediction models while retaining the same held-out calibration state.

```{r full-model-workflow, eval=FALSE, purl=FALSE, code=readLines(system.file("examples", "full-model-workflow.R", package = "vennabers"))}
```

## Shared multiclass full-model transfer {#full-multiclass-workflow}

`full-multiclass-workflow.R`: Use an additional full shared model with held-out OvO/OvR calibrators and inspect the retained state.

```{r full-multiclass-workflow, eval=FALSE, purl=FALSE, code=readLines(system.file("examples", "full-multiclass-workflow.R", package = "vennabers"))}
```

## Regression full-model transfer {#full-regression-workflow}

`full-regression-workflow.R`: Compare matched and full IVAR/CVAR workflows at fixed settings.

```{r full-regression-workflow, eval=FALSE, purl=FALSE, code=readLines(system.file("examples", "full-regression-workflow.R", package = "vennabers"))}
```

## Pairwise Inductive OvO training {#pairwise-inductive}

`pairwise-inductive.R`: Train a separate binary workflow for each class pair with matching Inductive calibration.

```{r pairwise-inductive, eval=FALSE, purl=FALSE, code=readLines(system.file("examples", "pairwise-inductive.R", package = "vennabers"))}
```

## Pairwise Cross OvO training {#pairwise-cross}

`pairwise-cross.R`: Retain the matching binary workflows and calibrators for every class pair and calibration fold.

```{r pairwise-cross, eval=FALSE, purl=FALSE, code=readLines(system.file("examples", "pairwise-cross.R", package = "vennabers"))}
```

## Calibrating a frozen binary workflow {#frozen-binary}

`frozen-binary.R`: Adopt a previously fitted recipe, model and postprocessor unchanged; compare losses on separate validation rows.

```{r frozen-binary, eval=FALSE, purl=FALSE, code=readLines(system.file("examples", "frozen-binary.R", package = "vennabers"))}
```

## Calibrating frozen multiclass and regression workflows {#frozen-multiclass-regression}

`frozen-multiclass-regression.R`: Separate earlier training and postprocessing from new VA calibration and independent assessment.

```{r frozen-multiclass-regression, eval=FALSE, purl=FALSE, code=readLines(system.file("examples", "frozen-multiclass-regression.R", package = "vennabers"))}
```

## Keeping the original schema for frozen adoption {#frozen-schema}

`frozen-schema.R`: Encode new calibration and assessment data using the schema retained before the original workflow training.

```{r frozen-schema, eval=FALSE, purl=FALSE, code=readLines(system.file("examples", "frozen-schema.R", package = "vennabers"))}
```