Hi everyone, I am trying to log the version of a versioned model that is used to make an inference and then add that alongside the inference output as either a column or a separate key. Do you know if this is something that is possible to do without a complicated workaround? For example, a model is trained and saved as a versioned dataset then in a different pipeline run the model is loaded to make an inference, and we would like to log that version timestamp name as a column in the predicted score table. The push is to add transparency to be able to note which model version has been used to make an inference score.

Hey @Ryan Ng there a few ways to approach this: • You could do a super simple subclass our

VersionedDataSet

or the underlying dataset your using to add the key automatically i.e.

InferenceParquetDataset

which automatically adds the extra column you describe above when used in your catalog. • You could do this on a more macro level using Kedro Hooks • You could also use Kedro’s native experiment tracking as an alternative workflow, but with minimal code changes.

Thanks for the suggestion! @datajoely - we're not able to use kedro viz at the minute so will explore the first 2 options for now 🙂