We are developing a custom evaluation library for our multi-class classification model. In our code repository, we utilized PySpark MLlib’s MulticlassMetrics for model evaluation. We have successfully committed our changes and created a tag for publication.
When configuring the evaluation dashboard, our custom evaluator is visible, and the “actual_field” can be selected.
Questions:
- Since our test data does not include an inference column for “inference_field”, should we manually add the expected model output name? After saving the configuration, will the model automatically run, generate inference results and add into our manually created output column?
- After configuration, our build failed with the following error:
Caused by: py4j.Py4JNetworkException: Error while sending a command: null response: c.
This appears to be a Py4J exception issue. We attempted to add Py4J version0.10.9.7to the requirements, and it passed the check. However, the environment still fails to initialize, and the build does not succeed. We then tried adding ‘openjdk’ and ‘java-jdk’ to the requirements, but the issue persists.
Any insights on resolving this would be greatly appreciated! 
