Overview

nmtvis is an open-source visualization toolkit designed for NMT(Neural Machine Translation) model.

This versatile and out-of-the-box toolkit aims at helping researchers better understand how their model works, so that they can further adjust or improve the model.

Installation

Use pip to install nmtvis::

  pip install nmtvis

Requirements

• Python3
• Numpy
• Scikit-learn

Features

• Visualize attention weights in attention-based NMT models
• Visualize high-dimensional vectors(e.g. word embeddings)
• Visualize the beam search decoding process

Links

• Source Code: https://github.com/player-eric/NMT-Visualizer
• Documentation: https://nmtvis.readthedocs.io/en/latest
• An article about the attention-visualization feature: http://player-eric.com/2020/02/20/nmtvis-attention/
• Attention mechanism visualization demo with example data: https://player-eric.github.io/attention_demo/
• Word embedding visualization demo with example data: https://player-eric.github.io/embedding_demo/
• Beam search decoding visualization demo with example data: https://player-eric.github.io/beam_demo/

License

The project is licensed under the MIT license.

Support

If you are having issues, please let me know. Contact me at digimonyan@gmail.com