README.md

metadata

title: Explaining Titanic
emoji: 🚢
short_description: Public multi-page ExplainerDashboard demo on Titanic models.
sdk: docker
app_port: 7860
tags:
  - python
  - docker
  - dash
  - explainable-ai

explainingtitanic

Demonstration of the explainerdashboard package.

This Dash app showcases model quality, permutation importances, SHAP values and interactions, individual trees, and multiple dashboard variants for sklearn-compatible models.

Live demos

• Heroku: https://titanicexplainer.herokuapp.com
• Fly.io (public proxy): https://titanicexplainer.fly.dev
• Fly.io (backend app): https://explainingtitanic.fly.dev
• Hugging Face Space: https://huggingface.co/spaces/oegedijk/explainingtitanic

ExplainerDashboard docs

• Docs: https://explainerdashboard.readthedocs.io
• Example notebook: https://github.com/oegedijk/explainerdashboard/blob/master/dashboard_examples.ipynb

Local run

With uv:

uv sync
uv run gunicorn --bind 0.0.0.0:7860 dashboard:app

Then open http://localhost:7860.

Hugging Face Spaces (Docker)

This repository is configured to run directly as a Hugging Face Docker Space.

• Runtime port: 7860
• Entrypoint: gunicorn dashboard:app
• Health endpoint: /healthz

If you duplicate this repo into a Space, set SDK to Docker and deploy.

Artifact strategy

This demo commits prebuilt explainer artifacts in pkls/*.joblib (about 8 MB total).

Why this choice:

• Faster and more reliable cold starts on free CPU Spaces.
• No extra model-building step during container startup.

Tradeoff:

• If model or sklearn versions change, regenerate artifacts with:

uv run python generate_explainers.py

Other deployment targets

• Fly.io: see FLY_DEPLOY.md.
• Heroku compatibility remains via Procfile and .heroku/run.sh.