ANN Optimization Dashboard (Streamlit)

Interactive Streamlit app for hyperparameter optimization of ANN regressors using Optuna. Upload data, tune the search space, run trials, and download trained models from the browser.

---

Features

• Web UI for loading tabular data, selecting features/targets, and setting train/test handling (split slider or separate test file).
• Optuna-powered search for ANN hyperparameters with live metrics (CV loss, R², NMAE).
• Parity plots and Optuna visualizations for completed runs.
• One-click downloads for models, predictions, and zipped run artifacts.
• Runs and config persisted to disk so you can reload and keep progress.

Prerequisites

• Python 3.10+
• Install dependencies:

  pip install -r requirements.txt

Run the app

streamlit run app.py

Open the URL shown in the terminal (usually http://localhost:8501).

Using the dashboard

1. Upload data
   • Upload a training file (csv, xlsx, or json).
   • Optional: click “Upload test data (optional)” to provide a separate test file; otherwise use the train/test split slider.
2. Select columns
   Choose feature columns (X) and a target column (Y) from the detected headers.
3. Configure training
   Adjust trials, learning rate/epochs, batch size, hidden layers/neurons, activation functions, optimizer, and CV settings. Enable feature standardization if needed.
4. Run + monitor
   Start the run from the control panel and watch live metrics/logs update.
5. Review results
   Inspect best metrics, parity plot, and Optuna charts, then download artifacts directly from the UI.

Configuration

• Working config lives in config.yaml; UI changes persist automatically and seed the next reload.
• Key sections:
  • variables: feature/target names.
  • cross_validation: split ratio, k-folds, and standardization toggle.
  • hyperparameter_search_space and network: bounds/choices for the Optuna search.
• You can edit config.yaml manually before launching if you prefer.

Outputs & folders

• runs/<dataset>__<status>: per-run folders with best model (.pt), predictions CSV, metrics, logs, and a zipped archive.
• optuna_study.db: Optuna study storage (SQLite).
• temp_data/: staging area for uploads (ignored by git).

Acknowledgements

Developed by the Institute for Dynamic Systems and Control at ETH Zürich.

License

MIT. See LICENSE for details.