GFI Problem

I have a problem with the “Global Feature Importance” node. In the metanode it seems to reset the names of the rows and then no longer be able to assign them correctly. Is there a solution?

Thanks!

My Workflow: https://daten-in-der.cloud/knime/Test.knwf

Did you find the solution?
br

Hi @ntq ,

thanks for reaching out. I had a look at the workflow that you attached. It differs from the one in the screenshot and produces different error:
ROC Curve (local) 0:2123:0:1735:0:1507:0:173:162 Sort column 'P (Prediction (Weight)=(7,20]) ' does not exist
This problem arose because in your training data you filtered out one category but didn’t update the domain. The component worked correctly after I added Domain Calculator after the Row Filter (Remove Inner Class) node.

I didn’t manage to reproduce the error from your screenshot. Would you like to share the workflow from the screenshot?

Lada

2 Likes

The correct workflow is now also available at the URL https://daten-in-der.cloud/knime/Test.knwf.

Hi @ntq ,

thank you for sharing. it seems that there is a bug in the AutoML component that causes the problem, not in the GFI component. We will fix it as soon as possible.

Since you can disconnect the component and fix or customize it, I will provide a quick workaround that you can use till we fix the problem:

  1. Right click on the AutoML component and click Component > Disconnect Link
  2. Open the AutoML component
  3. Open the AutoML Learner component
  4. Open the Generalized Linear Model (H2O) Meta Learner component
    4.1. In the Column Appender node select “Use the row keys from the table 1”
  5. Open the Generalized Linear Model (H2O) Meta Learner component
    5.1. In the Column Appender node select “Use the row keys from the table 1”

This should fix the bug for now. We will update you once we update the AutoML component on the KNIME Hub.

Let us know if you get more issues.

Lada

2 Likes

Thanks very much!

:smiley:

1 Like

This topic was automatically closed 7 days after the last reply. New replies are no longer allowed.