Machine Learning Inference with GitHub Actions

This repository demonstrated how you can use Github Actions to perform inference with your ML models. In this example scenario a Random Forest classifier is used to make predictions trained on the Iris dataset.

The GitHub Actions workflow is triggered when an issue receives a comment. If the comment contains the /predict prefix, then the main.py python file starts to parse the comment, make a prediction and construct the reply to the original comment with the prediction.

Example (and also valid) comment: /predict <sepal_length> <sepal_width> <petal_length> <petal_width> (e.g. /predict 5.6 2.9 3.6 1.3)

Try it out 😎

Just go to an issue at this repository and then leave a comment with the /predict prefic and then 4 numbers separated by spaces. Just like the example above.

/predict 5.6 2.9 3.6 1.3

Files

• action.yml: Describes the action which build a Docker image and performs the comment parsing and inference
• .github/workflows/main.yaml: contains the steps which are performed when a comment is received under an issue
• Dockerfile: This is the image which will be built and used for the main action
• issue_comment.sh: With this script you can send a comment with the GitHub Rest API
• main.py: parses the content of the comment, loads model, makes prediction and constructs the reply message
• random_forest_model.pkl: Serialized trained sklearn model which will be used for inference (btw. this file should not be here, as model artifacts should be stored in a storage, but this is just a sample so... 😄)

About

Gábor Vecsei

• Website & Blog
• LinkedIn
• Twitter
• Github