Maria Vechtomova's Linkedin Analytics

Before you deploy an MLflow model behind an endpoint on Databricks, Sagemaker, or AzureML, you need to make sure the endpoint behaves as expected. This is especially a concern when dealing with custom models, as MLflow has a complex logic for evaluating model input. You want to make sure you got that right. We propose the following steps when developing an MLflow custom model: - develop a custom model and log it in MLflow experiment tracking - download the artifact locally - serve the endpoint locally - run unit tests This is not something that is intended to run in a CI/CD pipeline, it is aimed to speed up the development process. Check out an article that I wrote together with Mehmet Acikgoz, Ph.D. for Decoding ML: https://lnkd.in/g88k5DN9. Thanks to Paul Iusztin for having us! #MLOps #machinelearning