xswem 1.0.0

A simple and explainable deep learning model for NLP.

XSWEM

A simple and explainable deep learning model for NLP implemented in TensorFlow.

Based on SWEM-max as proposed by Shen et al. in Baseline Needs More Love: On Simple Word-Embedding-Based Models and Associated Pooling Mechanisms, 2018.

This package is currently in development. The purpose of this package is to make it easy to train and explain SWEM-max.

You can find demos of the functionality we have implemented in the notebooks directory of the package. Each notebook has a badge that allows you to run it yourself in Google Colab. We will add more notebooks as new functionality is added.

For a demo of how to train a basic SWEM-max model see train_xswem.ipynb.

Local Explanations

We are currently implementing some methods we have developed for local explanations.

local_explain_most_salient_words

So far we have only implemented the local_explain_most_salient_words method. This method extracts the words the model has learnt as most salient from a given input sentence. Below we show an example of this method using a sample from the ag_news dataset. This method is explained in more detail in the local_explain_most_salient_words.ipynb notebook.

Global Explanations

We have implemented the global explainability method proposed in section 4.1.1 of the original paper. You can see a demo of this method in the notebook global_explain_embedding_components.ipynb.

How to install

This package is hosted on PyPI and can be installed using pip.

pip install xswem