Hands-On Deep Learning with R

Adding dropout layers

In the following code, we will add dropout layers and encourage you to experiment with different dropout rates to see what empirically leads to optimal results:

# initialize embedding parameterembedding_dim <- 64# create custom model with dropout layersdot_with_dropout <- function(  embedding_dim,  n_users,  n_items,  name = "dot_with_dropout") {  keras_model_custom(name = name, function(self) {    self$user_embedding <- layer_embedding(      input_dim = n_users+1,      output_dim = embedding_dim,      name = "user_embedding")    self$item_embedding <- layer_embedding(      input_dim = n_items+1,      output_dim = embedding_dim,      name = "item_embedding")    self$user_dropout <- layer_dropout(        rate = 0.2)    self$item_dropout <- layer_dropout(        rate = 0.4) self$dot ...