bavard-ml-common 0.0.3

Machine learning model utilities

bavard-ml-common

Web Service Example

import typing as t

from sklearn.datasets import load_iris
from sklearn.tree import DecisionTreeClassifier
import uvicorn

from bavard_ml_common.mlops.web_service import endpoint, WebService 


class DTModel(WebService):
    """
    A machine learning model that can turn into a web service.
    """

    def __init__(self) -> None:
        self._dt = DecisionTreeClassifier()
        self._fitted = False

    def fit(self, X, y) -> None:
        self._dt.fit(X, y)
        self._fitted = True

    # Type annotations are required for all `@endpoint` method arguments.
    @endpoint
    def predict(self, X: t.List[t.List[float]]):
        assert self._fitted
        return self._dt.predict(X).tolist()

    @endpoint
    def feature_importances(self) -> list:
        assert self._fitted
        return self._dt.feature_importances_.tolist()


# Fit a ML model on a dataset.
iris = load_iris()
X = iris.data
y = iris.target
model = DTModel()
model.fit(X, y)

# Convert the model into a `fastapi` web service.
api = model.to_app()

# Run the web service.
uvicorn.run(api, host="0.0.0.0", use_colors=True, log_level="debug")

# The fitted model is now available at 0.0.0.0:8000/,
# with a `/predict` endpoint and a `/feature_importances` endpoint.
# API documentation for the fitted model is at a `/docs` endpoint.

Developing Locally

Install dependencies:

pip3 install -e .
pip3 install -r requirements-test.txt

Then, run the tests using pytest:

python3 -m unittest

Releasing The Package

Releasing the package is automatically handled by CI, but three steps must be taken to trigger a successful release:

1. Increment the VERSION variable in setup.py to the new desired version (e.g. VERSION="1.1.1")
2. Commit and tag the repo with the exact same value you populated the VERSION variable with (e.g. git tag 1.1.1)
3. Push the commit and tag to remote. These can be done together using: git push --atomic origin <branch name> <tag>

CI will then release the package to pypi with that version once the commit and tag are pushed.