⚠️ Non-Functional: This project is under development.

This is an application to analyse the contribution of a text column to the performance of a machine learning model. The text data could be the only feature, or one column among many others. The application will provide local feature impact analysis to understand the specific qualities of the text (at word level) that are pushing model predictions either up or down.

To be released and distributed via setuptools/PyPI/pip for Python 3.

For Part of Speech Tagging we use spacy Note: After install you will need to get spaCy to download the English model.

sudo python3 -m spacy download en

We have built an internal synonym antonym dictionary from public access material. For an extended dictionary of synonyms and antonyms we use wordnet from NLTK You will need to install wordnet post-package-install.

import nltk
nltk.download('wordnet')

To use the application you will need a model that adheres to a simple interface It should go without saying that the model will need to utilise at least one text column. The model artefact should also contain any preprocessing required on the text. You will need a raw dataset containing text that you want to explain how it contributes to the model output.

You can use this application multiple ways

Use the runner without installing the application.

./textplainer-runner.py mymodel.pkl question data/test.csv > output.txt

Alternatively, you can invoke the directory as a package:

python -m textplainer 

Or simply install the package and use the command line application directly

Installation from the source tree:

python setup.py install

(or via pip from PyPI):

pip install textplainer

Now, the textplainer command is available::

textplainer mymodel.pkl question data/test.csv > output.txt

Python package built using the bootstrap cmdline template by jgehrcke