A Sustainability-Aware AutoML Knowledge Base

A lightweight Streamlit UI for exploring historical AutoML runs and getting intent-aware model recommendations from the AutoML knowledge base.

Features

• Explore datasets already stored in runs.db
• Compare candidates across frameworks such as H2O, AutoSklearn, and TPOT
• Visualize trade-offs between:
  • accuracy vs energy
  • accuracy vs latency
• Get conversational recommendations based on:
  • accuracy
  • energy
  • latency
• Apply optional constraints such as:
  • maximum inference latency
  • minimum accuracy
• Inspect the winning configuration and export it

Project structure

ICT4S26-demo/
├── automl_kb/
│   ├── config.py
│   └── data/
│       └── runs.db
└── apps/
    └── GUI/
        └── streamlit_app/
            └── app.py

Installation

From the project root:

python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

On Windows:

.venv\Scripts\activate
pip install -r requirements.txt

How to run

Run the app from the repository root so imports such as automl_kb.config resolve correctly:

streamlit run apps/GUI/streamlit_app/app.py

Then open the local Streamlit URL shown in the terminal.

Notes

• The app expects the SQLite database at automl_kb/data/runs.db
• Database paths are read from automl_kb/config.py
• If the DB is missing or the path is wrong, the app will show a connection error in the sidebar

Recommendation flow

1. Enter an OpenML task ID
2. Load the dataset if it exists in the knowledge base
3. Optionally view the framework leaderboard
4. Choose your optimization goal
5. Add constraints if needed
6. Review the recommended winner and alternatives
7. Inspect the configuration and export the result

Minimal dependencies

• streamlit
• pandas
• plotly

License