How to get run id from run name in MLflow - mlflow

Question

To download artifacts from a run, you need run id. I get the run id from the UI as shown below.
Run id from the UI
But when I set the run name parameter, run id is not visible in the UI. How to find the run Id of a particular run in MLflow ?

Answer 1

The run id in mlflow is a random stamp id generated. I had the same problem because I wrote an mlflow decorator, which needed access to the run id after the run was finished to set tags.
The question is what do you want to do after you have the run id? Then the approach would need extra infos.
If you only want to get access to your latest run:
Use the mlflow.list_run_infos() function and insert the experiment_id, which you can get by the mlflow.get_experiment_by_name function of mlflow. I guess you know your experiment's id. Here is the list_run_infos function
def list_run_infos(
self,
experiment_id: str,
run_view_type: int = ViewType.ACTIVE_ONLY,
max_results: int = SEARCH_MAX_RESULTS_DEFAULT,
order_by: Optional[List[str]] = None,
page_token: Optional[str] = None,
)
Then you should get a list of run objects. But, please read further:
In case you have multiple run objects in your experiment (that happens with several runs or even child runs from a parent run with Gridsearch and sklearn).
Loop through each! Runobject from the output of list__run_infos() and look at the end_time property of the Runobject. The endtime property is a UNIX timestamp. So even if you have a parent run or a single run, the highest UNIX timestamp in the end_time property will always be your last run (In case you didnt used several estimators in a loop in your experiment, that would require some refactoring). And by that you identify the appropriate RunObject.
Only then! you can access the Runtime Object's property: the run_id:
Here you can see the class of the run object from mlflow, keep in mind that you also need the exp_id again.
classmlflow.entities.RunInfo
(
run_uuid,
experiment_id,
user_id,
status,
start_time,
end_time,
lifecycle_stage,
artifact_uri=None,
run_id=None
)
In case you need the specific code:
last_parent_run = set()
exp_id = mlflow.get_experiment_by_name("your_exp_name"].experiment_id
for item in mlflow.list_run_infos(exp_id):
last_parent_run.add((item.__getattribute__("end_time"), item.__getattribute__("run_id")))
And then of course look for the maximum entry in your set
If you have any further questions just ask; i already tested my decorator with this and it works fine and keeps the main code of mlflow statements clean. Although its a little bit hacky, to access the run_id after the run.