Hi, I was experimenting a bit with the registratio...
# questions
Hi, I was experimenting a bit with the registration of metrics (both with
). It seems like both expect different data types as return values in the nodes, e.g.
wants a Dict
{"validation_loss" : validation_loss}
, whereas
wants a
Dict[str, Union[float, List[float]]]
to also specify the training step at which the metric is taken, e.g.
{ "validation_loss": {"value": validation_loss, "step": -1} }
. Now, what if I would like to keep track of the metrics in both tools? Is there a better way of doing this than having two different entries in the data catalog, i.e.
I plan to add the ``kedro-mlflow`` plugin to use the same syntax, but I am working on other things for now. I am afraid having 2 entries in the catalog is the simplest solution for now ( you coule override mlflow behaviour with a Hook or create a custom dataset too, but this is definitely not "simpler")
👍 1
@Joshua Beirer I faced exactly the same question and opened an issue in `kedro-mlflow`'s repo. See answer here And yes, having 2 catalog entries is the easiest workaround for this now.
👍 1