Tutorials¶
The tutorial notebooks walk through common OneEHR workflows with runnable examples.
| # | Tutorial | Covers |
|---|---|---|
| 1 | Quickstart | End-to-end TJH mortality prediction |
| 2 | Custom Dataset | Preparing dynamic, static, and label CSV files |
| 3 | Model Comparison | Comparing configured models with confidence intervals and statistical tests |
| 4 | Fairness And Explainability | Fairness metrics and feature importance outputs |
| 5 | Survival Analysis | DeepSurv, DeepHit, concordance index, and survival curves |
Run Locally¶
pip install oneehr jupyter
jupyter notebook tutorials/When working from a source checkout, install the package in editable mode first:
uv venv .venv --python 3.12
source .venv/bin/activate
uv pip install -e ".[test]"
jupyter notebook tutorials/Run In Colab¶
Upload a notebook to Google Colab and install OneEHR at the top:
!pip install oneehrWhat You Will Practice¶
Tutorial 1:
- Define an experiment with a TOML config file.
- Preprocess EHR events into binned features.
- Train XGBoost and GRU models.
- Evaluate on a held-out test split.
- Run comparison analysis.
Tutorial 2:
- Prepare dynamic, static, and label CSV files.
- Convert MIMIC-III, MIMIC-IV, or eICU data.
- Apply ICD, CCS, and ATC code mapping helpers.
Tutorial 3:
- Configure multiple models in one experiment.
- Read
predictions.parquetandmetrics.json. - Use bootstrap confidence intervals and pairwise statistical tests.
- Include configured LLM or agent systems in the same comparison table.
Tutorial 4:
- Compute fairness metrics across available static attributes.
- Review feature importance outputs.
- Connect analysis JSON files to plots and notebooks.
Tutorial 5:
- Configure DeepSurv and DeepHit.
- Train with survival task settings.
- Evaluate with concordance index.
- Plot survival curves from saved results.