def FeedForward(
inputs: str,
outputs: str,
units_inner: int,
units_readout: int,
dim: int,
dropout_rate: float,
):
"""FeedForward Layer."""
if inputs == "_x":
raise ValueError(
"Cannot use name '_x' for inputs (used as intermediary node).")
return Sequential(
Select(inputs=inputs, outputs="_x"),
Dropout(inputs="_x", outputs="_x", dropout_rate=dropout_rate),
Conv1d(inputs="_x",
outputs="_x",
filters=units_inner,
kernel_size=1,
activation=tf.nn.relu,
use_bias=True),
Dropout(inputs="_x", outputs="_x", dropout_rate=dropout_rate),
Conv1d(inputs="_x",
outputs="_x",
filters=units_readout,
kernel_size=1,
activation=None,
use_bias=True),
Dropout(inputs="_x", outputs="_x", dropout_rate=dropout_rate),
Dense(inputs="_x", outputs="_x", units=dim),
Add(inputs=(inputs, "_x"), outputs=outputs),
)
def CombineEmbeddings(tensors, mode, output_dim, project=True):
"""Combine Embeddings Layers"""
embedding = tf.concat(tensors, axis=-1)
if project:
embedding = Dense(units=output_dim)(embedding, mode=mode)
return embedding