Layer for positional encoding

Positional encoding layer with sine/cosine matrix of different frequencies.

Usage

layer_pos_sinusoid_wrapper(
  maxlen = 100,
  vocabulary_size = 4,
  n = 10000,
  load_r6 = FALSE,
  embed_dim = 64
)

Arguments

maxlen

Length of predictor sequence.

vocabulary_size

Number of unique character in vocabulary.

n

Frequency of sine waves for positional encoding. Only applied if pos_encoding = "sinusoid".

load_r6

Whether to load the R6 layer class.

embed_dim

Dimension for token embedding. No embedding if set to 0. Should be used when input is not one-hot encoded (integer sequence).

Value

A keras layer implementing positional encoding using sine/cosine waves.

Examples

if (FALSE) { # reticulate::py_module_available("tensorflow")
l <- layer_pos_sinusoid_wrapper() 
}