This code is meant to serve as an example of developing a pip-installable machine learning library that can be used for real-time inference by a deployed python package or another interoperable language.
This package demonstrates an ability to:
- Save model weights and include them in the python package for use after installation
- Use test driven development (TDD) to avoid scope creep and keep functions simple and maintainable
- Use node/pipeline architecture to simplify code execution when in production
- Use configs to handle variables that are likely to change over time for ease of finding and maintaining
- Design and implement a one-hot-encoding system that is reproducible even when specific categorical features are not present
After forking the repository and downloading locally:
Install build
pip install build
From the repo root, build wheel file
python -m build
.whl file will be built in the dist/
folder and can be installed by running:
pip install <path_to_whl>
>>> from example_package import Model
>>> model = Model()
>>> test = {
... "bedrooms": 4,
... "bathrooms": 2,
... "area": 7420,
... "basement": "no",
... "furnishingstatus": "furnished"
... }
>>> model.predict(test)
array([7094687.53284395])