Shapash : MAIF releases a new open source solution for a more transparent AI
MAIF, a french mutual insurance company (est. 1934), has been contributing to open-source for a few years. We released two solutions, Izanami and Otoroshi which are components of our micro-services platform. Since 2019, we also share our work in the field of artificial intelligence, our first release was Melusine, a tool for french langage email processing.
We are now releasing Shapash, a solution which aims to make AI algorithms more transparent and understandable . Thus, it contributes to the ethical use of data.
Elements of context
In France, a lot of people are suspicious with AI. This suspicion is increased by Big Data which provides more data which can be used to train AI algorithms.
This comes with several risks : misusage of private data, discriminatory biases … AI may also turn into unethical surveillance systems.
Within the European Union, public authorities are aware of these risks. They deal with these issues both at a European level with the GDPR and at a French level e.g. with the CNIL (French data protection agency).
Why we have made Shapash ?
At MAIF, as an insurer, we’ve been using data for a long time, to offer personalized advice to our customers as well as to control insurance risks. Because of our history, as a committed mutual insurance company and because we believe that companies should be responsible for their use of data, we are particularly involved in ethical usage of data.
We want our AI algorithms to be transparent because we believe it is essential to build and maintain trust in AI.
We have also been studying for a few years how Data science models can be more intelligible and understandable. This is a growing field of study, and several relevant open-source contributions have already been published about it. However, they mainly target data practitioners and specialists.
With Shapash, MAIF data scientists wanted to share a solution for everyone, no matter their data background.
At MAIF, we use Shapash to :
- Help data experts and non-experts to communicate better
- Make it easier and faster for MAIF’s data scientists to understand how models they develop work.
We believe that Shapash had to be open-sourced.
We want to share a solution for all, no matter their data background.
What is Shapash ?
Shapash a Python library that helps to make Machine Learning models more transparent and understood by everyone!
Concretely, it is an overlay to other intelligibility libraries (Shap, Lime) that:
- Displays understandable results with easy-to-read and simple wording and visualization, adapted to everyone.
- Displays an interface that helps to explore different features of a model and offers navigation between global and local explicability. This interface is particularly useful to animate workshops and answer questions on how ML models work.
- Summarizes local explicability to make it useful in an operational context. This summary is adaptable to different use cases and can be exported.
- Is open! Shapash is used for Regression, Classification, and fits a multitude of Machine Learning libraries, feature encoding (inverse encoding), usable with contributions calculated by Lime, Shap, …
Shapash can be used for all kinds of use cases : health, business, marketing, …
Please visit Shapash’s GitHub : you will there find more exhaustive documentation of the library’s features as well as a demo of the Web App Shapash Monitor for a quick test! And do not hesitate to give us a star on GitHub if you like the project!
