Opinionated machine learning pipelines for training and inference.
To do
To do.
Declare any dependencies in pyproject.toml using uv add [package-name].
This will also create a virtual environment .venv if not exists and installs the package.
To create a virtual environment and install all dependencies, run:
uv sync --frozen
Add a --no-dev flag to omit installation of dev dependencies.
You can run your Kedro project with:
uv run kedro run
Candidate options: xgboost, lightgbm and catboost. We compared the three packages based on 3 criteria:
- number of releases in 2022-2023
- number of dependencies (important when working with containers)
- package size
Based on those criteria, we had the following result:
| xgboost | lightgbm | catboost | |
|---|---|---|---|
| # releases | 14 | 7 | 9 |
| dependencies | 2 (numpy/scipy) | 2 (numpy/scipy) | 18 |
| size (MB) | 6.2 | 5.1 | 90 |