Package {densemlp}

Title: Dense Neural Networks for Tabular Classification and Regression
Version: 0.5.0
Author: Imad EL BADISY [aut, cre]
Maintainer: Imad EL BADISY <elbadisyimad@gmail.com>
Description: Provides dense feed-forward neural network models for tabular regression and classification using 'torch'. The package supports modern extensions around dense neural network blocks, including dropout, batch normalization, residual connections, gated blocks, and optional input projection.
URL: https://github.com/ielbadisy/densemlp
BugReports: https://github.com/ielbadisy/densemlp/issues
License: MIT + file LICENSE
Encoding: UTF-8
RoxygenNote: 7.3.3
Imports: ggplot2, stats, torch
Suggests: knitr, rmarkdown, testthat (≥ 3.0.0)
Config/testthat/edition: 3
VignetteBuilder: knitr
NeedsCompilation: no
Packaged: 2026-08-03 18:54:34 UTC; imad-el-badisy
Repository: CRAN
Date/Publication: 2026-08-08 13:50:11 UTC

densemlp: Dense Multilayer Perceptrons for Tabular Data

Description

Provides dense feed-forward multilayer perceptron models for tabular regression and classification. Models are built from fully connected neural network layers, with optional dropout, batch normalization, residual connections, gated blocks, and input projection.

Details

The models implemented in densemlp are dense feed-forward multilayer perceptrons. Optional components such as dropout, batch normalization, residual connections, gated blocks, and input projection extend the basic dense multilayer perceptron architecture but do not change the model class into a convolutional, recurrent, transformer, or tree-based model.

Author(s)

Maintainer: Imad EL BADISY elbadisyimad@gmail.com

See Also

Useful links:

Plot training history

Description

Plot training history

Usage

## S3 method for class 'densemlp_fit'
autoplot(object, ...)

Arguments

object

A fitted densemlp_fit object.

...

Unused.

Value

A ggplot object.

Fit a tabular dense multilayer perceptron

Description

Fit a tabular dense multilayer perceptron

Usage

densemlp(
  formula = NULL,
  data = NULL,
  x = NULL,
  y = NULL,
  task = c("auto", "classification", "regression"),
  hidden_units = c(64, 32),
  activation = c("relu", "tanh", "gelu"),
  dropout = 0,
  batch_norm = TRUE,
  residual = FALSE,
  gated = FALSE,
  input_projection = NULL,
  epochs = 100,
  batch_size = 32,
  lr = 0.001,
  optimizer = c("adam", "sgd"),
  lr_schedule = c("none", "cosine", "step"),
  weight_decay = 0,
  validation = 0.2,
  early_stopping = TRUE,
  patience = 10,
  min_delta = 0,
  min_epochs = max(10L, floor(epochs * 0.2)),
  loss = NULL,
  label_smoothing = 0,
  focal_gamma = 2,
  metrics = NULL,
  seed = 1,
  verbose = TRUE,
  log_every = 1,
  device = c("auto", "cpu", "cuda")
)

Arguments

formula

A formula specification.

data

A data frame used with formula.

x

Predictor data frame or matrix. Retained for backward compatibility with the x/y interface.

y

Outcome vector. Retained for backward compatibility with the x/y interface.

task

Optional task override. "auto" infers the task from the outcome.

hidden_units

Hidden layer sizes.

activation

Activation function.

dropout

Dropout probability.

batch_norm

Use batch normalization in hidden layers.

residual

Use residual skip connections between hidden blocks.

gated

Use learned gating inside hidden blocks.

input_projection

Optional input projection dimension before hidden blocks.

epochs

Number of epochs.

batch_size

Mini-batch size.

lr

Learning rate.

optimizer

Optimizer name.

lr_schedule

Learning-rate schedule.

weight_decay

Weight decay.

validation

Validation fraction.

early_stopping

Enable early stopping.

patience

Early stopping patience.

min_delta

Minimum validation loss improvement.

min_epochs

Minimum number of epochs before early stopping can trigger.

loss

Loss function. "focal" is available for binary classification.

label_smoothing

Label smoothing for classification losses.

focal_gamma

Focal-loss focusing parameter.

metrics

Reserved for future custom metrics.

seed

Random seed.

verbose

Verbosity level. FALSE or 0 silences output, TRUE or 1 prints a standard log, and 2 prints a detailed log.

log_every

Epoch logging frequency.

device

Device to use.

Value

A densemlp_fit object.

Compute densemlp metrics

Description

Compute densemlp metrics

Usage

densemlp_metrics(truth, estimate, task = NULL, prob = NULL)

Arguments

truth

Ground truth values.

estimate

Predicted classes for classification or numeric predictions for regression.

task

Optional task override. When NULL, the task is inferred from truth and estimate.

prob

Optional class probabilities.

Value

A named list of metrics.

Permutation variable importance

Description

Permutation variable importance

Usage

perm_importance(object, new_data, truth, metric = NULL, seed = object$seed)

Arguments

object

A fitted densemlp_fit object.

new_data

Evaluation data frame.

truth

Ground truth outcome values.

metric

Metric name. Defaults to accuracy for classification and RMSE for regression.

seed

Random seed used for shuffling.

Value

A densemlp_importance object.

Plot permutation importance

Description

Plot permutation importance

Usage

## S3 method for class 'densemlp_importance'
plot(x, ...)

Arguments

x

A densemlp_importance object.

...

Unused.

Value

A ggplot object.

Plot training history

Description

Plot training history

Usage

plot_history(object)

Arguments

object

A fitted densemlp_fit object.

Value

A ggplot object.

Predict from a fitted dense multilayer perceptron

Description

Predict from a fitted dense multilayer perceptron

Usage

## S3 method for class 'densemlp_fit'
predict(object, new_data, type = NULL, ...)

Arguments

object

A fitted densemlp_fit object.

new_data

New predictor data.

type

Prediction type.

...

Unused.

Value

Predictions in a task-appropriate format.

Print a fitted dense multilayer perceptron

Description

Print a fitted dense multilayer perceptron

Usage

## S3 method for class 'densemlp_fit'
print(x, ...)

Arguments

x

A fitted densemlp_fit object.

...

Unused.

Value

x, invisibly.

Tune a dense multilayer perceptron over a task-aware hyperparameter grid

Description

Tune a dense multilayer perceptron over a task-aware hyperparameter grid

Usage

tune_densemlp(
  formula = NULL,
  data = NULL,
  x = NULL,
  y = NULL,
  task = c("auto", "classification", "regression"),
  grid = NULL,
  metric = NULL,
  validation = 0.2,
  early_stopping = TRUE,
  patience = 10,
  min_delta = 0,
  min_epochs = NULL,
  seed = 1,
  repeats = 3,
  verbose = FALSE,
  device = c("auto", "cpu", "cuda"),
  refit = TRUE
)

Arguments

formula

A formula specification.

data

A data frame used with formula.

x

Predictor data frame or matrix.

y

Outcome vector.

task

Optional task override. "auto" infers the task from the outcome.

grid

A named list of candidate values.

metric

Optional ranking metric. Use "accuracy" for classification and "rmse" or "valid_loss" for regression.

validation

Validation fraction for each fit.

early_stopping

Enable early stopping during tuning.

patience

Early stopping patience.

min_delta

Minimum validation loss improvement.

min_epochs

Minimum number of epochs before early stopping can trigger.

seed

Base random seed.

repeats

Number of repeated seeds per candidate.

verbose

Print per-candidate progress.

device

Device to use.

refit

Refit the best configuration on the supplied data.

Value

A list with ranked tuning results and, when refit = TRUE, the best fitted model.