A Python utility for command-line interaction with Azure Machine Learning model endpoints. I built this as a testing interface for a flight delay prediction model trained with Azure's AutoML.
How it works
AutoML handled feature engineering, algorithm selection, and hyperparameter tuning. I wrote a CLI tool to send prediction requests to the deployed endpoint with flight parameters (origin, destination, departure, arrival times) as arguments.
Results
- R² Score: 0.80
- Normalized RMSE: Under 0.1
Solid first-pass results. The model captured most of the variance in the data and made consistent predictions across the range of values.
What I learned
Compared to building a model from scratch, AutoML trades granular control for speed. It tested multiple algorithms and configurations in parallel and handled the infrastructure, so no GPU driver headaches like I had on Lambda Labs.
Tech stack
- Azure ML Studio with AutoML — feature engineering and model search
- Python with the Requests library — CLI client for the deployed endpoint