Evaluation Dataset Schema Is Not A Subset Of The Trained Model Schema

News Knowledge Base
, , ,
1

How to troubleshoot evaluation dataset schema is not a subset of the trained model schema?

Issue:

When running a full pipeline, this error typically occurs when the schema of the dataset being used to evaluate the model is different than the schema of the model.

Validation:

For the Input Dataset used and the Evaluation Dataset used, open the schema for each dataset and compare to ensure there are no differences in field names etc.

  1. Click on the failed pipeline and make a note of the Input Dataset and the Evaluation Dataset selected for the Pipeline's configuration. (It is helpful to take a screenshot of what was used for reference) Example screenshot with paths
  1. Navigate to Datasets and open the full path that is being used for the input directory. Download the schema.json
  1. Navigate back out to Datasets and open the full path of the Evaluation directory that was used. Download the schema.json
  1. Open both schemas and do a side-by-side comparison. (Tip: In Notepad++ a plugin called "Compare" is available. By using the plugin you can compare two files to spot any differences.)

Note: The schemas should be the same, if any field names are different, for example, invoice-number in the training dataset schema but invoice_number in the evaluation dataset schema, the error would occur.

Resolution:

After the differences are discovered between the schemas, Open the data labeling session for the training dataset or the evaluation dataset and change the field name so that it matches the other schema. Export the dataset from Document Manager and use the new exported dataset to run a new full pipeline.

Next Steps:

After following the steps as recommended above, if the issue is still occurring, share the following with UiPath Product Support:

  1. Is the issue occurring in UiPath Cloud AiCenter, On-Prem Standalone AI Center, or Automation Suite Deployed AI Center?
  2. Capture the ML Pipeline Details
    1. Review the following Guide for how to capture a HAR trace - How To Generate A HAR File.
    2. Open UiPath Cloud AiCenter
    3. Go to the Network tab in Dev Tools(F12) in the browser
    4. Navigate to the Pipelines page
    5. Click on the specific pipeline that failed
    6. Save the HAR trace to share with UiPath Product Support.

1 Like