kindling: '{kindling}': Higher-level interface of torch package to...

In kindling: Higher-Level Interface of 'torch' Package to Auto-Train Neural Networks

{kindling}: Higher-level interface of torch package to auto-train neural networks

Description

{kindling} enables R users to build and train deep neural networks such as:

• Deep Neural Networks / (Deep) Feedforward Neural Networks (DNN / FFNN)

• Recurrent Neural Networks (RNN)

It is mainly designed to generate code expressions of the current architectures which happens to reduce boilerplate {torch} code for the said current architectures. It also integrate seamlessly with titanic ML frameworks - currently with {tidymodels}, which enables components like {parsnip}, {recipes}, and {workflows}, allowing ergonomic interface for model specification, training, and evaluation.

Thus, the package supports hyperparameter tuning for:

• Number of hidden layers

• Number of units per layer

• Choice of activation functions

Clarification: The hyperparameter tuning support is supported in version 0.1.0, but n_hlayer() dial parameter is not supported. This changes after 0.2.0 release.

Details

The {kindling} package provides a unified, high-level interface that bridges the {torch} and {tidymodels} ecosystems, making it easy to define, train, and tune deep learning models using the familiar tidymodels workflow.

How to use

The following uses of this package has 3 levels:

Level 1: Code generation

ffnn_generator(
    nn_name = "MyFFNN",
    hd_neurons = c(64, 32, 16),
    no_x = 10,
    no_y = 1,
    activations = 'relu'
)

Level 2: Direct Execution

ffnn(
    Species ~ .,
    data = iris,
    hidden_neurons = c(128, 64, 32),
    activations = 'relu',
    loss = "cross_entropy",
    epochs = 100
)

Level 3: Conventional tidymodels interface

# library(parsnip)
# library(kindling)
box::use(
   kindling[mlp_kindling, rnn_kindling, act_funs, args],
   parsnip[fit, augment],
   yardstick[metrics],
   mlbench[Ionosphere] # data(Ionosphere, package = "mlbench")
)

# Remove V2 as it's all zeros
ionosphere_data = Ionosphere[, -2]

# MLP example
mlp_kindling(
    mode = "classification",
    hidden_neurons = c(128, 64),
    activations = act_funs(relu, softshrink = args(lambd = 0.5)),
    epochs = 100
) |>
    fit(Class ~ ., data = ionosphere_data) |>
    augment(new_data = ionosphere_data) |>
    metrics(truth = Class, estimate = .pred_class)
#> A tibble: 2 × 3
#>   .metric  .estimator .estimate
#>   <chr>    <chr>          <dbl>
#> 1 accuracy binary         0.989
#> 2 kap      binary         0.975

# RNN example (toy usage on non-sequential data)
rnn_kindling(
    mode = "classification",
    hidden_neurons = c(128, 64),
    activations = act_funs(relu, elu),
    epochs = 100,
    rnn_type = "gru"
) |>
    fit(Class ~ ., data = ionosphere_data) |>
    augment(new_data = ionosphere_data) |>
    metrics(truth = Class, estimate = .pred_class)
#> A tibble: 2 × 3
#>   .metric  .estimator .estimate
#>   <chr>    <chr>          <dbl>
#> 1 accuracy binary         0.641
#> 2 kap      binary         0

Main Features

• Code generation of {torch} expression

• Multiple architectures available: feedforward networks (MLP/DNN/FFNN) and recurrent variants (RNN, LSTM, GRU)

• Native support for {tidymodels} workflows and pipelines

• Fine-grained control over network depth, layer sizes, and activation functions

• GPU acceleration supports via {torch} tensors

License

MIT + file LICENSE

Author(s)

Maintainer: Joshua Marie [e-mail]

Authors:

• Antoine Soetewey [e-mail] (ORCID)

References

Falbel D, Luraschi J (2023). torch: Tensors and Neural Networks with 'GPU' Acceleration. R package version 0.13.0, https://torch.mlverse.org, https://github.com/mlverse/torch.

Wickham H (2019). Advanced R, 2nd edition. Chapman and Hall/CRC. ISBN 978-0815384571, https://adv-r.hadley.nz/.

Goodfellow I, Bengio Y, Courville A (2016). Deep Learning. MIT Press. https://www.deeplearningbook.org/.

See Also

Useful links:

• https://kindling.joshuamarie.com

• https://github.com/joshuamarie/kindling

• Report bugs at https://github.com/joshuamarie/kindling/issues

kindling documentation built on March 3, 2026, 9:07 a.m.