Open astrojuanlu opened 1 month ago
This is definitely a feature, not a bug 😉 even if I acknowledge the error message could be better (but I did not imagine someone would ever do that!)
Essentially what happens is that :
6ffe573f95124a65a04dcece1589fbdc
mlflow_run_id
run because of
# mlflow.yml
tracking:
run:
id: ${runtime_params:mlflow_run_id}
This is very unlikely that you want to modify an existing run, but say you want to complete it
However, I assume you want to load the regressor you've just trained. This is perfectly possible:
# catalog.yml
regressor:
type: kedro_mlflow.io.models.MlflowModelTrackingDataset
flavor: mlflow.sklearn
run_id: ${runtime_params:mlflow_run_id}
# note that you don't specify anything in mlflow yml regarding the run id, you will either create a new mlflow run (or don't create a run if you added it to the blacklist)
and now you can run:
kedro run --pipeline inference --params mlflow_run_id=xxx
(note that this is not the training pipeline, but a one that consumes the model)
(On a side note, it is likely a bad idea to deal specify the run id manually on each run, just specify a registered_model_name
and let the model registry handle it for you)
Does it make sense?
My understanding from the beginning was that this is a feature 😄 The point was more around the error message.
In any case, thanks a bunch for the explanation 🙏🏼 it makes a ton of sense.
I leave it to you whether you can to keep this issue open for improving the error message, or just close it as won't fix.
(On a side note, it is likely a bad idea to deal specify the run id manually on each run, just specify a
registered_model_name
and let the model registry handle it for you)
Description
I read in the docs that
...but I'm stubborn so I still wanted to try.
For that, I specified my
run_id
both in the configuration and in the dataset:And when launching
kedro run --params mlflow_run_id=xxx
I get the following error:Expected Result
If the
active_run
and the specifiedrun_id
are the same, proceed.Actual Result
(The error above)
Your Environment
kedro
andkedro-mlflow
version used (pip show kedro
andpip show kedro-mlflow
): 0.19.5 and 0.12.2 respectivelypython -V
): 3.11.5Does the bug also happen with the last version on master?
Yes, it still does.