I found it tricky to display SHAP (static and interactive) plots in Streamlit.
Those of you who’ve tried in the past have likely had to write custom wrappers to display interactive SHAP plots (that use JS) in Streamlit apps, and reuse the wrapper code across multiple projects. I realized I could save you the effort by packaging it into a Streamlit Component
LMK what you think! Contributions are more than welcome
This is great @snehankekre - do you have any examples of using this with DNNs?
@asehmi Not at the moment. Great suggestion though. I will add a couple in the coming days.
In the meantime, feel free to have to look at any DNN examples in the SHAP repo. Adapting them to Streamlit should as simple as wrapping the final
shap. call in the examples with
st_shap() (provided the linked examples aren’t outdated/broken).
That would be great if you have time. Thanks, @snehankekre .
I’ve added two deep learning examples in v1.0.2. They’re heavily based on examples from the SHAP repo.
@snehankekre Amazing. I’ll have a look. Thanks a million.