For mlflow autologging we currently log and trace methods like 'fit', 'predict', 'score', 'transform'.
Via Events we get information when certain substeps start, complete and extra attributes, like
- Staging
- Preprocessing
- Checking
- Feature trying
- Feature building
- Trainging
- …
Use mlflow spans to trace those information under the parent trace.
https://mlflow.org/docs/latest/tracking.html
https://mlflow.org/docs/latest/python_api/mlflow.client.html#mlflow.client.MlflowClient.start_span
https://mlflow.org/docs/latest/python_api/mlflow.client.html#mlflow.client.MlflowClient.end_span
https://github.com/getml/monorepo/pull/370