| title | Explaining Titanic | ||||
|---|---|---|---|---|---|
| emoji | 🚢 | ||||
| short_description | Public multi-page ExplainerDashboard demo on Titanic models. | ||||
| sdk | docker | ||||
| app_port | 7860 | ||||
| tags |
|
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.
- 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
- Docs: https://explainerdashboard.readthedocs.io
- Example notebook: https://github.com/oegedijk/explainerdashboard/blob/master/dashboard_examples.ipynb
With uv:
uv sync
uv run gunicorn --bind 0.0.0.0:7860 dashboard:appThen open http://localhost:7860.
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.
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- Fly.io: see
FLY_DEPLOY.md. - Heroku compatibility remains via
Procfileand.heroku/run.sh.