Goal of this repo is to test out the real-time and batch endpoints as part of Azure ML. Future state could be to understand how auto-scaling triggers work, and if logs are properly reported. Borrowed the source Jupyter notebook from here. Main difference is the addition of the batch deployment, and some logic to generate slightly larger inference json and csv payloads to test the endpoints. For example, when passing a file with a 1 million records, the real-time endpoint fails and this is ideal for a batch endpoint to process this volume.
- Setup the virtual environment locally. Install needed dependencies using the
make installcommand. - Trigger the creation of resources using the
make infracommand. - Use the steps in the Jupyter notebook:
e2e-ml-workflow.ipynbto follow the sequence of steps to create data assets, build a training pipeline, setup both real-time and batch endpoints, and then test it with generated data (off the original dataset).