Hi everyone, I’ve started to use

kedro-mlflow

and am under the impression that there is some conflict / incompatibility with the dataset factory. Having moved from

"models.{model}.embeddings":
    type: pickle.PickleDataSet
    filepath: "data/07_model_output/{model}_embeddings.pkl"
    backend: joblib
    versioned: True
    layer: model_output

to

"models.{model}.embeddings":
  type: kedro_mlflow.io.artifacts.MlflowArtifactDataSet
  data_set:
    type: pickle.PickleDataSet
    filepath: "data/07_model_output/{model}_embeddings.pkl"
    backend: joblib
    versioned: True
  layer: model_output

results in artifacts being saved without the model named properly interpolated, i.e “litteraly” saved as

{model}_embeddings.pkl

Any comment ? Anything I’m missing ? Many thanks in advance, Regards, M.

hi @Marc Gris! just to make sure, are you using the

OmegaConfigLoader

and the latest version of Kedro?

It should work with any config loader

I see, this is a bug on our end

dataset factories isn’t specific to OCL

When we resolve the config it goes through each entry and fills out the placeholder, but here the filepath is nested

Hi everyone Thanks for your swift reactions / replies. I’ve tried both with 0.18.12 & 0.18.13 and get the same issue. Just to be extra clear: The dataset factory patterns works fine, until I use

<http://kedro_mlflow.io|kedro_mlflow.io>.artifacts.MlflowArtifactDataSet

I hope this is a trivial bug to fix 🙂 @Ankita Katiyar Could you please share the ticket here, so that I may “follow” it and be automatically informed when it is fixed ? Many thanks in advance, Regards M.

Here’s the issue https://github.com/kedro-org/kedro/issues/2992 🙂 The fix should be simple will open a PR for it

Thx @Ankita Katiyar 🙂 🙏🏼

Hey @Marc Gris, we’re dealing with CI issues at the moment 😅 but it’s ready to go!

Great ! Thx @Ankita Katiyar 🙂

@Marc Gris it’s merged!

🥳 Thx @Ankita Katiyar