Duplicate run names intentional?

[ablaom]

When I tried out the example and repeated the evaluation to generate new runs, they all had the same name. Is that intentional?

[pebeto]

Yes, it is intentional because it was programmed to give you a hint of what kind of model belongs to that run. However, if it's needed, we can remove that naming and allow mlflow to generate a random name (not related to anything about the run).

[ablaom]

Okay, I'm assuming there's nothing the user could want to do that actually requires them to identify a run by name, and so this acceptable. It Just seems a bit weird for run names to be duplicated. Generally, I think of "names" as synonymous with identifiers. Better would be "DecisionTreeClassifier 1", "DecisionTreeClassifier 2", and so forth, but I'm assuming that's a lot more complicated to implement, because you need a counter, and so forth. Or, how about DecisionTreeClassifier …1342, DecisionTreeClassifier …3857, etc, where 1342 is the last four digits of the run id?

Alternatively, we let MLflow generate it's random names and add the model type name as a tag instead.

What do you think?