Skip to contents

During training, randomly zeroes some of the elements of the input tensor with probability p using samples from a Bernoulli distribution. Each channel will be zeroed out independently on every forward call.


nn_dropout(p = 0.5, inplace = FALSE)



probability of an element to be zeroed. Default: 0.5


If set to TRUE, will do this operation in-place. Default: FALSE.


This has proven to be an effective technique for regularization and preventing the co-adaptation of neurons as described in the paper Improving neural networks by preventing co-adaptation of feature detectors.

Furthermore, the outputs are scaled by a factor of :math:\frac{1}{1-p} during training. This means that during evaluation the module simply computes an identity function.


  • Input: \((*)\). Input can be of any shape

  • Output: \((*)\). Output is of the same shape as input


if (torch_is_installed()) {
m <- nn_dropout(p = 0.2)
input <- torch_randn(20, 16)
output <- m(input)