Package {tabby}

Title: Tidy Interfaces for Tabular Deep Learning
Version: 0.0.1
Description: Model definitions for fitting, tuning, and predicting with deep learning models for tabular data using the 'parsnip' and 'tidymodels' frameworks. Supported models include the 'TabPFN' foundation model of Hollmann et al. (2025) <doi:10.1038/s41586-024-08328-6>, 'TabICL' by Qu et al. (2025) <doi:10.48550/arXiv.2502.05564>, 'SAINT' by Somepalli et al. (2021) <doi:10.48550/arXiv.2106.01342>, and the tabular 'ResNet' of Gorishniy et al. (2021) <doi:10.48550/arXiv.2106.11959>, among others. Helper functions are also included to create space-filling grids for tuning parameters that are defined on a per-layer basis. These functions define the model interfaces; the computations are carried out by separate engine packages.
License: MIT + file LICENSE
URL: https://github.com/tidymodels/tabby, https://tabby.tidymodels.org/
BugReports: https://github.com/tidymodels/tabby/issues
Depends: parsnip (≥ 1.6.1), R (≥ 4.1.0)
Imports: cli, dials (≥ 1.4.4), dplyr, generics, hardhat, purrr, rlang (≥ 1.1.0), tibble
Suggests: brulee (≥ 1.2.0), modeldata, spelling, testthat (≥ 3.0.0), torch, withr
Encoding: UTF-8
Language: en-US
Config/Needs/website: tidyverse/tidytemplate, brulee
Config/testthat/edition: 3
Config/roxygen2/version: 8.1.0
NeedsCompilation: no
Packaged: 2026-10-05 12:26:23 UTC; max
Author: Max Kuhn ORCID iD [aut, cre], Edgar Ruiz [aut], Posit Software, PBC ROR ID [cph, fnd]
Maintainer: Max Kuhn <max@posit.co>
Repository: CRAN
Date/Publication: 2026-10-15 17:00:02 UTC

tabby: Tidy Interfaces for Tabular Deep Learning

Description

logo

Model definitions for fitting, tuning, and predicting with deep learning models for tabular data using the 'parsnip' and 'tidymodels' frameworks. Supported models include the TabPFN foundation model of Hollmann et al. (2025) doi:10.1038/s41586-024-08328-6, TabICL by Qu et al. (2025) doi:10.48550/arXiv.2502.05564, SAINT by Somepalli et al. (2021) doi:10.48550/arXiv.2106.01342, and the tabular ResNet of Gorishniy et al. (2021) doi:10.48550/arXiv.2106.11959, among others. Helper functions are also included to create space-filling grids for tuning parameters that are defined on a per-layer basis. These functions define the model interfaces; the computations are carried out by separate engine packages built on 'torch'.

Author(s)

Maintainer: Max Kuhn max@posit.co (ORCID)

Authors:

Other contributors:

See Also

Useful links:

Model predictions across many sub-models

Description

Model predictions across many sub-models

Usage

## S3 method for class ''_brulee_auto_int''
multi_predict(object, new_data, type = NULL, epochs = NULL, ...)

## S3 method for class ''_brulee_resnet''
multi_predict(object, new_data, type = NULL, epochs = NULL, ...)

## S3 method for class ''_brulee_rln''
multi_predict(object, new_data, type = NULL, epochs = NULL, ...)

Arguments

object

A model fit.

new_data

A rectangular data object, such as a data frame.

type

A single character value or NULL. Possible values are:

  • regression: "numeric"

  • classification: "class", "prob"

  • censored regression: "survival", "time", "hazard", "linear_pred"

  • quantile regression: "quantile"

  • interval estimates: "conf_int", "pred_int"

  • other: "raw"

When NULL, predict() will choose an appropriate value based on the model's mode.

epochs

An integer vector for the number of training epochs.

...

Optional arguments to pass to predict.model_fit(type = "raw") such as type.

Value

A tibble with the same number of rows as new_data. Its .pred column is a list of tibbles, each containing the predictions for the corresponding row of new_data across the requested values of epochs.

Working with Tuning Parameters that are Lists

Description

neural_net_grid_space_filling() takes the number of layers in a neural network as an input, creates a space-filling design with the multidimensional layer parameter(s), and collapses the grid into list parameters that the model function needs. expand_list_parameters() is a convenience function that converts the list columns to a wide structure for printing, visualization, etc.

Usage

neural_net_grid_space_filling(
  wflow,
  num_layers = 3,
  size = 10,
  param_info = NULL,
  range = c(2, 50),
  collapse = TRUE
)

expand_list_parameters(x, pattern = "*")

Arguments

wflow

A model specification or workflows::workflow() object.

num_layers

A single integer for the number of layers containing hidden units.

size

The minimum grid size. It may be adjusted upwards to find a feasible design.

param_info

A dials::parameters() object or NULL. If none is given, a parameters set is derived from other arguments. Passing this argument can be useful when parameter ranges need to be customized.

range

A range for the number of hidden units in each layer.

collapse

A single logical for whether to collapse the parameters into list columns.

x

A data frame of grid points, some of which are list columns.

pattern

A regular expression pattern to select which list columns should be expanded to a wide format.

Value

neural_net_grid_space_filling() returns a tibble with grid points, some of which are list-columns containing integer vectors. expand_list_parameters() returns a tibble where the selected list columns are replaced by one integer column per layer.

Examples

rn_spec <-
  tabular_resnet(hidden_units = tune(),
             bottleneck_units = tune(),
             penalty = tune())

rn_grid <- neural_net_grid_space_filling(rn_spec)
rn_grid

rn_grid |> expand_list_parameters()

Objects exported from other packages

Description

These objects are imported from other packages. Follow the links below to see their documentation.

generics

required_pkgs()

parsnip

check_args(), multi_predict()

mirror server hosted at Truenetwork, Russian Federation.