The flavour MLflow uses to carry arbitrary Python, whether code or a model, into Model Serving. Custom models extend PythonModel, then go through mlflow.pyfunc.log_model.
Also called Python function.
Read more: Databricks docs
In the Ultra Transcenders books
Each book explains pyfunc in context, with comparison tables and the common traps.
Terms in this definition
- MLflow
An open-source framework for experiment tracking and model management. Azure Machine Learning adopts it as its tracking API, so each workspace acts as a compatible MLflow tracking server.
- Model Serving
A Databricks service for deploying, governing and querying AI models, in real time or in batches. Each served model becomes a REST API on serverless compute, and custom agents hosted there now carry a legacy label.
Related terms
- MLflow flavours
Alternative formats,
sklearnandpyfuncfor instance, that MLflow can store one model in so that various later tools can load it. Dependencies on packages are recorded for you with native ones. - pandas UDF
Also called a vectorised UDF. You declare this Python function with
pandas_udfand type hints; Apache Arrow moves the data, pandas does the work, and rows arrive in batches instead of singly.