Keras Attention Layer

Version (s)

TensorFlow: 1.15.0 (Tested)
TensorFlow: 2.0 (Should be easily portable as all the backend functions are availalbe in TF 2.0. However has not been tested yet.)

Introduction

This is an implementation of Attention (only supports Bahdanau Attention right now)

Project structure

data (Download data and place it here)
 |--- small_vocab_en.txt
 |--- small_vocab_fr.txt
layers
 |--- attention.py (Attention implementation)
examples
 |--- nmt
   |--- model.py (NMT model defined with Attention)
   |--- train.py ( Code for training/inferring/plotting attention with NMT model)
   |--- train_variable_length_seq.py ( Code for training/inferring with variable length sequences)
 |--- nmt_bidirectional
   |--- model.py (NMT birectional model defined with Attention)
   |--- train.py ( Code for training/inferring/plotting attention with NMT model)
models (created by train_nmt.py to store model)
results (created by train_nmt.py to store model)

How to use

Just like you would use any other tensoflow.python.keras.layers object.

from attention_keras.layers.attention import AttentionLayer

attn_layer = AttentionLayer(name='attention_layer')
attn_out, attn_states = attn_layer([encoder_outputs, decoder_outputs])

Here,

encoder_outputs - Sequence of encoder ouptputs returned by the RNN/LSTM/GRU (i.e. with return_sequences=True)
decoder_outputs - The above for the decoder
attn_out - Output context vector sequence for the decoder. This is to be concat with the output of decoder (refer model/nmt.py for more details)
attn_states - Energy values if you like to generate the heat map of attention (refer model.train_nmt.py for usage)

Visualizing Attention weights

An example of attention weights can be seen in model.train_nmt.py

After the model trained attention result should look like below.

Running the NMT example

Prerequisites

In order to run the example you need to download small_vocab_en.txt and small_vocab_fr.txt from Udacity deep learning repository and place them in the data folder.

Using the docker image

If you would like to run this in the docker environment, simply running run.sh will take you inside the docker container.
Set the GPU_TAG in the run.sh appropriately depending on whether you need the GPU version of the CPU version

Using a virtual environment

If you would like to use a virtual environment, first create and activate the virtual environment.
Then, use either
- pip install -r requirements.txt -r requirements_tf_cpu.txt (For CPU)
- pip install -r requirements.txt -r requirements_tf_gpu.txt (For GPU)

Running the code

Go to the . Any example you run, you should run from the folder (the main folder). Otherwise, you will run into problems with finding/writing data.
Run python3 src/examples/nmt/train.py. Set degug=True if you need to run simple and faster.
If run successfully, you should have models saved in the model dir and attention.png in the results dir.

If you have improvements (e.g. other attention mechanisms), contributions are welcome!

Dipeshpal / attention_keras