kedro-mlflow

question: My pipeline can train models for different clients, these are deployed separately. If i want to use kedro-mlflow to train and deploy all the models in one go using the

pipeline_ml_factory

(see: https://kedro-mlflow.readthedocs.io/en/stable/source/05_pipeline_serving/04_hook_pipeline_ml.html), should i make sure to train one model per kedro run? or can i have multiple

pipeline_ml_factory

, and filter on two sets of tags (“training” and “client_name” )?

your

register_pipelines

can register several pipelines @Hugo Evers, it would be a matter of calling

pipeline_ml_factory

more than once if that's what you want. cc'ing @Yolan Honoré-Rougé in case I'm missing something

yea, i was thinking about how the prediction pipelines are isolated. so that could be on a run basis (in which case id specify a global to switch between client models) or on a pipeline basis, so every pipeline can only train one model.