keras-pos-embd 0.13.0

Position embedding layers in Keras

Keras Position Embedding

[中文|English]

Position embedding layers in Keras.

Install

pip install keras-pos-embd

Usage

Trainable Embedding

from tensorflow import keras
from keras_pos_embd import PositionEmbedding

model = keras.models.Sequential()
model.add(PositionEmbedding(
    input_shape=(None,),
    input_dim=10,     # The maximum absolute value of positions.
    output_dim=2,     # The dimension of embeddings.
    mask_zero=10000,  # The index that presents padding (because `0` will be used in relative positioning).
    mode=PositionEmbedding.MODE_EXPAND,
))
model.compile('adam', 'mse')
model.summary()

Note that you don't need to enable mask_zero if you want to add/concatenate other layers like word embeddings with masks:

from tensorflow import keras
from keras_pos_embd import PositionEmbedding

model = keras.models.Sequential()
model.add(keras.layers.Embedding(
    input_shape=(None,),
    input_dim=10,
    output_dim=5,
    mask_zero=True,
))
model.add(PositionEmbedding(
    input_dim=100,
    output_dim=5,
    mode=PositionEmbedding.MODE_ADD,
))
model.compile('adam', 'mse')
model.summary()

Sin & Cos Embedding

The sine and cosine embedding has no trainable weights. The layer has three modes, it works just like PositionEmbedding in expand mode:

from tensorflow import keras
from keras_pos_embd import TrigPosEmbedding

model = keras.models.Sequential()
model.add(TrigPosEmbedding(
    input_shape=(None,),
    output_dim=30,                      # The dimension of embeddings.
    mode=TrigPosEmbedding.MODE_EXPAND,  # Use `expand` mode
))
model.compile('adam', 'mse')
model.summary()

If you want to add this embedding to existed embedding, then there is no need to add a position input in add mode:

from tensorflow import keras
from keras_pos_embd import TrigPosEmbedding

model = keras.models.Sequential()
model.add(keras.layers.Embedding(
    input_shape=(None,),
    input_dim=10,
    output_dim=5,
    mask_zero=True,
))
model.add(TrigPosEmbedding(
    output_dim=5,
    mode=TrigPosEmbedding.MODE_ADD,
))
model.compile('adam', 'mse')
model.summary()