Option for using auto-generated names for kedro runs instead of pipeline name

[AlexandreOuellet]

Description

When running the same kedro pipeline multiple time in parallel (doing experiments in parallel), I would like an option to generate random name for those runs, instead of always having __default__ or pipelinename

Context

In my scenario, I have a lot of parameters to test, and I want to test them in parallel (1 kedro run per parameters), and I don't really care about an experiment name (although I do want to be able to refer to them quickly). Instead of spending time naming my runs pipelinename_flipr and coming up with names and/or exposing the values there, I would like to simply have the run name be automatically generated (so that I can still refer to it in the future if needed).

Possible Implementation

Have a setting in mlflow.tml's tracking->run to specify wether to use a predefined name (like it is currently) or to have a randomly generated run name.

Possible Alternatives

Right now, I would need to modify mlflow.yml's tracking->run->name parameter for each run for each parameter values I would like to modify.

[Galileo-Galilei]

Hi, I agree such an option would be great! I thought that explictly writing:

server: 
    tracking:
        run:
            name: null

but it uses __default__ by default. I need to think about the best syntax since I guess this is the best default but I agree kedro-mlflow should support your use case.

[Galileo-Galilei]

Note for myself: I have found a way to make this possible and consistent with the interactive workflow (which is the hardest part): I'll create a custom resolver which will import the _generate_string from mlflow.

I'll add this after 0.19 is officically out.

[MosaicMan]

I would love to see this merged. Any way I can help?

[Galileo-Galilei]

I would love to see this merged. Any way I can help?

Sorry for the late reply. I had little time recently + there is a very subtle bug in how kedro handles resolvers that needs to be addressed on the core framework side, but I release this feature since the bug should very likely be hardly noticeable, and this is a commonly required feature.

[MosaicMan]

Thank you so much for all your great work into this package!

[Galileo-Galilei]

Thanks!
This is already on PyPI, you can try with kedro-mlflow==0.12.1

For the record the syntax is :

#mlflow.yml
tracking: 
    run: 
        name: ${km.random_name:} # don't forget the trailing ":" at the end ! 

[MosaicMan]

It is a beautiful thing. Thank you!