Setup mlflow configuration when context is loaded

[Galileo-Galilei]

Description

When I call load_context() interactively, the configuration of the mlflow.yml is:

• not accessible through a context attributes
• not set up (especially, the mlflow tracking uri is not set)

Context

I always struggle when i want to use the configuration in the mlflow.yml interactively. This makes experimentation & debugging harder.

Possible Implementation

For now, I can't see how it is possible to modify the context object to add attributes. We can eventually add some to the catalog with the after_catalog_created hook, but the configuration should not be tied to the catalog execution.

[takikadiri]

I understand the point, but to my knowledge the kedro context do not manage plugins properties, it just use them to supercharge his own properties (catalog, params, pipelines)

It isn't sufficient for you to get the config by calling the get_mlflow_config of kedro_mlflow ?

context = load_context(Path.cwd())
mlflow_config = get_mlflow_config(context)

[Galileo-Galilei]

The current implementation seems indeed satisfying enough (with kedro>=0.17 and the a KedroSession activated, it can be simplified to

from kedro_mlflow.framework.context
mlflow_config=get_mlflow_config()
mlflow_config.setup()

However, we should setup automatically the configuration for jupyter, through the %reload_kedro line magic.