Setting the mlflow experiment does not work in interactive mode

[Galileo-Galilei]

Description

If I specify an experiment in mlflow.yml, and the set up the mlflow configuration interactively, all runs should be stored by default in this experiment while they are currently sotred in mlflow "Default" (0) experiment. This works when running "kedro run" through the CLI.

Steps to Reproduce

# mlflow.yml
experiment:
  name: my_awesome_experiment
  create: True  # if the specified `name` does not exists, should it be created?

# test.py

from kedro.framework.session import KedroSession
from kedro.framework.startup import bootstrap_project
from kedro_mlflow.config import get_mlflow_config

bootstrap_project(r"path/to/project")
with KedroSession.create(project_path=r"path/to/project"):
    config=get_mlflow_config()
    config.setup()
    
    mlflow.log_param("test_param",1) # this should be logged in "my_awesome_experiment" but is logged in "Default".

Does the bug also happen with the last version on master?

Yes

Potential solution

The faulty line is:

kedro-mlflow/kedro_mlflow/config/kedro_mlflow_config.py

Line 100 in 904207a

self._get_or_create_experiment()

We should use mlflow mlflow.set_experiment method, but it does not restore deleted experiment. This wil replace part of the logic here:

kedro-mlflow/kedro_mlflow/config/kedro_mlflow_config.py

Lines 124 to 132 in 904207a

	if self.experiment.create:
	if self._experiment is None:
	# case 1 : the experiment does not exist, it must be created manually
	experiment_id = self._mlflow_client.create_experiment(
	name=self.experiment.name
	)
	self._experiment = self._mlflow_client.get_experiment(
	experiment_id=experiment_id
	)