layer_to_dense_DeepTRIAGE: Apply a DeepTRIAGE Layer
In tpq/caress: Caress, a Gentler Introduction to Keras

This function applies a variant of the DeepTRIAGE attention mechanism to the incoming layer (see DOI:10.1101/533406). This implementation differs slightly from the publication in that all layers have the same activation function and the random embedding weights are optionally learnable.

layer_to_dense_DeepTRIAGE(
  object,
  result_dim,
  embed_dim = result_dim * 4,
  random_embedding = FALSE,
  hidden_dim = 32,
  hidden_activation = "tanh",
  hidden_dropout = 0.5,
  name = NULL
)

`object`	A `keras` model.
`result_dim`	An integer. The size of the final layer.
`embed_dim`	An integer. The size of the final embedding matrix will equal the input dimension times the embedding dimension.
`random_embedding`	A boolean. Toggles whether to freeze the embedding matrix with random values. Otherwise, the embedding matrix is trainable.
`hidden_dim`	An integer. The size of the hidden layers.
`hidden_activation`	A string. The activation for the hidden layers.
`hidden_dropout`	A numeric. The dropout for the hidden layers.
`name`	A string. The prefix label for all layers.