Retrieve feature importance from partioned model

Options
julesbertrand
julesbertrand Partner, Dataiku DSS Core Designer, Dataiku DSS ML Practitioner, Dataiku DSS Adv Designer, Registered Posts: 9 Partner

Hello,

Is there a way to automatically retrieve (via a recipe) features importances of a partitioned model, by partition, in DSS please ?

I tried to do it through a python recipe with the following code:

import dataikuimport pandas as pdclient = dataiku.api_client()project = client.get_project(dataiku.default_project_key())model = project.get_saved_model("BE8E0mVs")perf = model.get_version_details('initial').get_raw().get("iperf")

But then perf is empty, I cannot do perf.retrieve("rawImportance") as suggested here. The code snippet works on an unpartitioned model though.

Thanks

Tagged:

Best Answer

  • Alexandru
    Alexandru Dataiker, Dataiku DSS Core Designer, Dataiku DSS ML Practitioner, Dataiku DSS Adv Designer, Registered Posts: 1,209 Dataiker
    Answer ✓
    Options

    Hi @julesbertrand
    ,

    The get_predictor() and get_version_metrics() are not yet supported for partitioned models. We do have this feature request in our backlog.

    Thanks,

Answers

  • June
    June Dataiku DSS Core Designer, Registered Posts: 19 ✭✭✭✭
    Options

    I'd like to upvote the backlog enhancement for support of partitioned models in the API. Share the link to the enhancement request?

Setup Info
    Tags
      Help me…