Transformer model for tabular longitudinal data.
The ukko codebase is a Python package for deep learning on tabular longitudinal (time-series) and survival data, with a focus on transformer-based and Cox proportional hazards models.
- Several survival heads (survival model output layers) are beeing explored atm
- Key feature of teh core models is a dual attention mechanism
ukko/
├── pyproject.toml
├── src/
│ └── ukko/
│ ├── __init__.py
│ └── core.py
└── ukko_get_started.ipynb-
pyproject.toml:
- Configures the project metadata (name, version, author, description, etc.).
- Uses hatchling as the build backend.
- Specifies the source code is under src/ukko.
- Includes build instructions for both wheel and sdist distributions.
-
src/ukko/:
- This is the main package directory.
__init__.py: Initializes the package and imports everything from core.py.core.py: Function definitions of the package.ukko_get_started.ipynb: A Jupyter notebook containing usage examplesfor the ukko package.
-
Editable Install:
The project is set up for development with editable installs (pip install -e .), making it easy to test changes. -
Modern Packaging:
Uses pyproject.toml and the src layout, which is recommended for modern Python projects. -
Intended Usage:
The package is designed for transformer models applied to tabular longitudinal data, as described in the project metadata.