SuperLearner wrapper for neural network using torch

A SuperLearner wrapper that implements a neural network using torch, with a softplus activation to encourage strictly positive predictions. By default, the neural network uses a single hidden layer with 10 hidden units, an Adam optimizer with a learning rate of 0.01, an MSE loss function, and 100 epochs.

Usage

SL.nnet.torch.softplus(
  Y,
  X,
  newX,
  n_hidden = 10,
  learning_rate = 0.01,
  loss_fn = torch::nn_mse_loss(),
  epochs = 100,
  lower = 1e-04,
  ...
)

Arguments

Y

A numeric vector of outcome values.

X

A numeric matrix or data.frame of covariates and treatment.

newX

A numeric matrix or data.frame of predictors.

n_hidden

Number of hidden units in the hidden layer, default is 10.

learning_rate

Learning rate for the Adam optimizer, default is 0.01.

loss_fn

The loss function used during training, default is MSE.

epochs

The number of epochs for training, default is 100.

lower

Lower bound to clip predictions to, default is 1e-4

...

Any additional arguments.

Value

A list with components:

• pred: A numeric vector of predictions on newX.

• fit: A list containing the fitted model object.