Hi! Could you provide some examples how to use mlflow logging? Readme description doesn't help me.
I launch server using mlflow server --backend-store-uri mlflow_server. As a train script using train_dual.py and at the top I import mflow, set mlflow.set_registry_uri(uri = 'http://127.0.0.1:5000'), and set
experiment_id = mlflow.create_experiment(
"test",
artifact_location=Path.cwd().joinpath("mlruns").as_uri(),
tags={"version": "v1", "priority": "P1"},
)
but I don't see anything in the web UI. Probably I'm wrong somewhere

@nikiniki1
As explained in the README, you only have to define the different environment variables (MLFLOW_TRACKING_URI, MLFLOW_EXPERIMENT_NAME, MLFLOW_RUN) before launching your script. Tracking will then be done automatically. You don't need to import mlflow, this will also be done automatically.

More information on MLflow environment variables available: https://mlflow.org/docs/latest/python_api/mlflow.environment_variables.html

Currently, it is not possible to directly record tags using environment variables. This could be an interesting feature to add (on the mlflow github directly)

Don't hesitate to tell me if it works for you.

@Youho99
So, I set these env variables and get message WARNING mlflow.utils.autologging_utils: You are using an unsupported version of pytorch. If you encounter errors during autologging, try upgrading / downgrading pytorch to a supported version, or try upgrading MLflow. I found my new experiment and run, but when I click 'model metrics' I didn't see anything. Torch=2.2.1, mlflow = 2.11.1

@nikiniki1
The warning is normal

The logging seems to work correctly since the run is created, but the tracking of metrics poses a problem...

Open an issue on my fork, so that we can discuss it more deeply and find the solution to your problem.

@nikiniki1 As explained in the README, you only have to define the different environment variables (MLFLOW_TRACKING_URI, MLFLOW_EXPERIMENT_NAME, MLFLOW_RUN) before launching your script. Tracking will then be done automatically. You don't need to import mlflow, this will also be done automatically.

More information on MLflow environment variables available: https://mlflow.org/docs/latest/python_api/mlflow.environment_variables.html

Currently, it is not possible to directly record tags using environment variables. This could be an interesting feature to add (on the mlflow github directly)

Don't hesitate to tell me if it works for you.

Hi! I use docker create mlflow web server
and then i change the mlflow/init.py line 30 to uri = os.environ.get("MLFLOW_TRACKING_URI", "http://localhost:5000/")
but it seems nothing happened in my mlflow web UI
is that the right way to use ? thanks