Base models for Neural Network Training in kindling

Usage

ffnn(
  formula,
  data,
  hidden_neurons,
  activations = NULL,
  output_activation = NULL,
  bias = TRUE,
  epochs = 100,
  batch_size = 32,
  learn_rate = 0.001,
  optimizer = "adam",
  loss = "mse",
  validation_split = 0,
  device = NULL,
  verbose = FALSE,
  cache_weights = FALSE,
  ...
)

rnn(
  formula,
  data,
  hidden_neurons,
  rnn_type = "lstm",
  activations = NULL,
  output_activation = NULL,
  bias = TRUE,
  bidirectional = TRUE,
  dropout = 0,
  epochs = 100,
  batch_size = 32,
  learn_rate = 0.001,
  optimizer = "adam",
  loss = "mse",
  validation_split = 0,
  device = NULL,
  verbose = FALSE,
  ...
)

Arguments

formula: Formula. Model formula (e.g., y ~ x1 + x2).
data: Data frame. Training data.
hidden_neurons: Integer vector. Number of neurons in each hidden layer.
activations: Activation function specifications. See act_funs().
output_activation: Optional. Activation for output layer.
bias: Logical. Use bias weights. Default TRUE.
epochs: Integer. Number of training epochs. Default 100.
batch_size: Integer. Batch size for training. Default 32.
learn_rate: Numeric. Learning rate for optimizer. Default 0.001.
optimizer: Character. Optimizer type ("adam", "sgd", "rmsprop"). Default "adam".
loss: Character. Loss function ("mse", "mae", "cross_entropy", "bce"). Default "mse".
validation_split: Numeric. Proportion of data for validation (0-1). Default 0.
device: Character. Device to use ("cpu", "cuda", "mps"). Default NULL (auto-detect).
verbose: Logical. Print training progress. Default FALSE.
cache_weights: Logical. Cache weight matrices for faster variable importance computation. Default FALSE. When TRUE, weight matrices are extracted and stored in the returned object, avoiding repeated extraction during importance calculations. Only enable if you plan to compute variable importance multiple times.
...: Additional arguments passed to the optimizer.
rnn_type: Character. Type of RNN ("rnn", "lstm", "gru"). Default "lstm".
bidirectional: Logical. Use bidirectional RNN. Default TRUE.
dropout: Numeric. Dropout rate between layers. Default 0.

Value

An object of class "ffnn_fit" containing:

model: Trained torch module
formula: Model formula
fitted.values: Fitted values on training data
loss_history: Training loss per epoch
val_loss_history: Validation loss per epoch (if validation_split > 0)
n_epochs: Number of epochs trained
feature_names: Names of predictor variables
response_name: Name of response variable
device: Device used for training
cached_weights: Weight matrices (only if cache_weights = TRUE)

FFNN

Train a feed-forward neural network using the torch package.

RNN

Train a recurrent neural network using the torch package.

Examples

if (FALSE) { # \dontrun{
# Regression task (auto-detect GPU)
model_reg = ffnn(
    Sepal.Length ~ .,
    data = iris[, 1:4],
    hidden_neurons = c(64, 32),
    activations = "relu",
    epochs = 50,
    verbose = FALSE
)

# With weight caching for multiple importance calculations
model_cached = ffnn(
    Species ~ .,
    data = iris,
    hidden_neurons = c(128, 64, 32),
    activations = "relu",
    cache_weights = TRUE,
    epochs = 100
)
} # }

if (FALSE) { # \dontrun{
# Regression with LSTM on GPU
model_rnn = rnn(
    Sepal.Length ~ .,
    data = iris[, 1:4],
    hidden_neurons = c(64, 32),
    rnn_type = "lstm",
    activations = "relu",
    epochs = 50
)

# With weight caching
model_cached = rnn(
    Species ~ .,
    data = iris,
    hidden_neurons = c(128, 64),
    rnn_type = "gru",
    cache_weights = TRUE,
    epochs = 100
)
} # }