How to handle errors in MLflow when a model has been served using "mlflow models serve"?

Question

During training, it is possible to use tags as a way to handle exceptions according to this question.

If a model has been created using mlflow.pyfunc.PythonModel, is it possible to throw exceptions? Is there a way to allow error handling for a model that has been served?

score 0 · Answer 1 · answered Aug 04 '20 at 17:39

0

One way is to to catch exception in your overridden predict() method for the mlflow.pyfunc.PyModel and log it.

answered Aug 04 '20 at 17:39

Jules

44
2

Since we expose the model as a REST API when we serve it, what should be returned in case of exception? – Aditya Khadse Aug 05 '20 at 05:39

How to handle errors in MLflow when a model has been served using "mlflow models serve"?

1 Answers1