I just wanted to add some info for future reference in case there’s interest. There were coincidental LinkedIn posts this past week profiling each of:
Both of them:
- are self contained libraries that deliver an ML model explainability app experience
- leverage python plotly dash underneath (i.e. in my layperson’s head, I might think of them as essentially super sophisticated all-in-one components)
- have lots of github stars, generate lots of responses when profiled on LI
One of the SHAPASH developers (from MAIF - Thomas Bouché) helped provide me with some info r.e. compare/contrast of the two:
explainerdashboard is a great library that came out in the same period as Shapash. This shows the need to have an app to visualize the explainability of machine learning models.
*The Shapash Webapp is built to have an UX that allows you to easily navigate between local and global explainability. The app is also designed to have non-technical labels and be able to present it to non-users of the data. *
In addition to the Webapp, Shapash can also generate a report to have a basis for an audit document of the machine learning model.
*There are also quality metrics of explainability to increase the confidence that the data scientist can have in explainability. And the possibility to compile Shap, Lime, ACV or to give your own contributions. *
explainerdashboard has additional features, on the metrics of the model, the concept of what if, and others.