Attention mechanism for processing sequence data that considers the context for each timestamp.
Install
pip install keras-self-attentionUsage
Basic
By default, the attention layer uses additive attention and considers the whole context while calculating the relevance. The following code creates an attention layer that follows the equations in the first section (attention_activation is the activation function of e_{t, t'}):
import keras
from keras_self_attention import Attention
model = keras.models.Sequential()
model.add(keras.layers.Embedding(input_dim=10000,
output_dim=300,
mask_zero=True))
model.add(keras.layers.Bidirectional(keras.layers.LSTM(units=128,
return_sequences=True)))
model.add(Attention(attention_activation='sigmoid'))
model.add(keras.layers.Dense(units=5))
model.compile(
optimizer='adam',
loss='categorical_crossentropy',
metrics=['categorical_accuracy'],
)
model.summary()Local Attention
The global context may be too broad for one piece of data. The parameter attention_width controls the width of the local context:
from keras_self_attention import Attention
Attention(
attention_width=15,
attention_activation='sigmoid',
name='Attention',
)Multiplicative Attention
You can use multiplicative attention by setting attention_type:
from keras_self_attention import Attention
Attention(
attention_width=15,
attention_type=Attention.ATTENTION_TYPE_MUL,
attention_activation=None,
kernel_regularizer=keras.regularizers.l2(1e-6),
use_attention_bias=False,
name='Attention',
)Regularizer
To use the regularizer, the attention should be returned for calculating loss:
import keras
from keras_self_attention import Attention
inputs = keras.layers.Input(shape=(None,))
embd = keras.layers.Embedding(input_dim=32,
output_dim=16,
mask_zero=True)(inputs)
lstm = keras.layers.Bidirectional(keras.layers.LSTM(units=16,
return_sequences=True))(embd)
att, weights = Attention(return_attention=True,
attention_width=5,
attention_type=Attention.ATTENTION_TYPE_MUL,
kernel_regularizer=keras.regularizers.l2(1e-4),
bias_regularizer=keras.regularizers.l1(1e-4),
name='Attention')(lstm)
dense = keras.layers.Dense(units=5, name='Dense')(att)
model = keras.models.Model(inputs=inputs, outputs=[dense, weights])
model.compile(
optimizer='adam',
loss={'Dense': 'sparse_categorical_crossentropy', 'Attention': Attention.loss(1e-2)},
metrics={'Dense': 'categorical_accuracy'},
)
model.summary(line_length=100)
model.fit(
x=x,
y=[
numpy.zeros((batch_size, sentence_len, 1)),
numpy.zeros((batch_size, sentence_len, sentence_len))
],
epochs=10,
)Load the Model
Make sure to add Attention to custom objects and add attention_regularizer as well if the regularizer has been used:
import keras
keras.models.load_model(model_path, custom_objects={
'Attention': Attention,
'attention_regularizer': Attention.loss(1e-2),
})1 maintainer
Metadata
1 maintainer
Release files for keras-self-attention 0.0.16
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | Provenance | Actions |
|---|---|---|---|---|
| keras-self-attention-0.0.16.tar.gz | 5.2 kB | Without provenance | Details for keras-self-attention-0.0.16.tar.gz |
Release files / keras-self-attention-0.0.16.tar.gz
| Download URL | keras-self-attention-0.0.16.tar.gz |
|---|---|
| Size | 5.2 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
e6c3a51fe29b88477454e233d1acc1cd6a22fc58bc3583fcfe14e3787590e584
|
|
BLAKE2b-256 checksum How to use checksums |
774403ddfa115da1db2875526c47cb1f48a4785fcc484248e16730fab06678ae
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/1.11.0 pkginfo/1.4.2 requests/2.18.4 setuptools/28.8.0 requests-toolbelt/0.8.0 tqdm/4.24.0 CPython/3.6.4
|
Security for keras-self-attention 0.0.16
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
No provenance found
None of the files in this release were uploaded with provenance metadata.
A note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.
To learn more about protecting yourself from malicious actors, visit our package security documentation.
