How to add attention layer to seq2seq model on Keras

Question

Based on this article, I wrote this model:

enc_in=Input(shape=(None,in_alphabet_len))
lstm=LSTM(lstm_dim,return_sequences=True,return_state=True,use_bias=False)
enc_out,h,c=lstm(enc_in)
dec_in=Input(shape=(None,in_alphabet_len))
decoder,_,_=LSTM(decoder_dim,return_sequences=True,return_state=True)(dec_in,initial_state=[h,c])
decoder=Dense(units=in_alphabet_len,activation='softmax')(decoder)
model=Model([enc_in,dec_in],decoder) 

How can I add attention layer to this model before decoder?

Inanimate

here a simple way to add attention: https://mcmap.net/q/586055/-how-to-add-attention-layer-to-a-bi-lstm – Roncesvalles

Answer 1

You can use this repo,

1. you will need to pip install keras-self-attention
2. import layer from keras_self_attention import SeqSelfAttention
   • if you want to use tf.keras not keras, add the following before the import os.environ['TF_KERAS'] = '1'
   • Make sure if you are using keras to omit the previous flag as it will cause inconsistencies

3. Since you are using keras functional API,

   enc_out, h, c = lstm()(enc_in)
   att = SeqSelfAttention()(enc_out)
   dec_in = Input(shape=(None, in_alphabet_len))(att)
   

   I hope this answers your question, and future readers

Freddie

Well this is self attention. Yet, for seq2seq you will normally want to have attention between encoder and decoder states. – Femur

So, what do you suggest? – Freddie

I don't know if there is a Keras wrapper for Bahdanau or Luong attention, yet there is a neat TensorFlow 2.0 tutorial for seq2seq translation with attention. tensorflow.org/tutorials/text/nmt_with_attention – Femur