Cloud Run ML API Demo

This repo contains a minimal example of using Github to deploy a machine learning model to Google Cloud Run. It uses ONNX to save and load the model and falcon to serve it.

Example

import pprint
import requests

payload = {
    "sepal_length_cm": 5.1,
    "sepal_width_cm": 3.5,
    "petal_length_cm": 1.4,
    "petal_width_cm": 0.2
}

resp = requests.post("https://my-cloud-run-url/predict", json=payload)
pprint.pprint(resp.json())

Outputs:

{'predictions': [{'class_name': 'Iris Setosa',
                  'class_no': 0,
                  'score': 0.9766749739646912},
                 {'class_name': 'Iris Versicolour',
                  'class_no': 1,
                  'score': 0.023324977606534958},
                 {'class_name': 'Iris Virginica',
                  'class_no': 2,
                  'score': 6.551524478481952e-08}]}

QA

• Tox is used to orchestrate the tests.
• The dev-requirements.txt file contains useful libraries for managing the code.
• A pre-commit config is provided to help enforce code quality tooling.

Improvements

There are several improvements that can be considered, for example:

• Logging, using libraries such as structlog.
• Making sure that the raw scores from the ONNX model are actual probabilities.
• Better tests and documentation (forever).