An MLOps Workflow

This exercise focuses on applying some Kubernetes concepts to an MLOps workflow.

The data science team in your company uses an algorithm to detect unusually large values (outliers) in a time series from an IoT device. This algorithm is constantly improved, leading to frequent changes that end up as new versions of a respective Docker image. Your task is to support them by putting these algorithms into production, helping the data science team to test the individual performance of the versions.

The following has been prepared:

• The actual data to run the detection algorithm against (data.csv), which will be copied into a Docker image
• Two different versions of the outlier detection algorithm (detect_outliers_*.py)
• Two Dockerfiles for creating two images containing these different versions (Dockerfile.outlier_detection_*)
• A bash script to prepare and upload the images (01_images.sh)
• A Kubernetes Manifest called 02_pods_outlier-detection.yml is available, which you will use to deploy the outlier detection algorithms.