Troubleshooting Machine learning issues

Use the following topics to resolve machine learning issues in the Persona-based UI.

Table 1. Log locations
Log Location
ML logs $TOP/mdmui/logs/machinelearning
Scheduler and Workflow logs $TOP/logs
Containerized deployment
Following is the command to get into a pod.
 oc rsh <POD_NAME>
Admin logs /logs/admin_<ADMIN_POD_NAME>
ML logs /logs/ml_<ML_POD_NAME>
Scheduler logs /logs/scheduler_<SHEDULER_POD_NAME>
Workflow logs /logs/workflowengine_<WORKFLOW_POD_NAME>

Machine Learning Training Report job returns a blank report.out file

Symptoms
Machine Learning Training Report job returns a blank report.out file.
Causes
  • ML services are in stopped state.
  • Machine Learning Training Report is unable to read the training the Microsoft Excel.
  • The application is unable to establish a MongoDB connection.
Solution
  • Check the scheduler and then ML log of the service for which you have triggered training.
  • Check the status of ML services and start the services.
  • Check whether a valid Microsoft Excel training is present on the configured path.
  • Check whether the MongoDB details that are configured in the $TOP/bin/conf/env_settings.ini file are correct. Ensure that you have enabled MongoDB authentication. For more information, see Enabling MongoDB authentication.

ML training has started, but the status of the model is In progress

Symptoms
ML training has started, but the status of the model is In progress.
Causes
Memory and CPU configuration is insufficient and hence it is taking more time to complete the training.
Solution
  • Check the ML log of the service for which you have triggered training.
  • Increase the memory and CPU configuration, and restart the training.

Error while saving a script in a workflow step

Symptoms
You get an error while saving a script in a workflow step.
Causes
  • Script that is provided in the script editor contains some error.
  • The machinelearning-ext.jar file is not configured properly.
Solution
  • Check the ipm.log file of the Admin service.
  • Correct the script. For more information, see Configuring machine learning Services for valid script for a step.
    Note: If any customization is done in the machinelearning-ext.jar, then script should be provided according to the customization.
  • Check whether the JAR is present on the server and its full path is present in the class path.

Not able to get prediction while processing an item

Symptoms
You are not getting any prediction while processing an item through a collaboration area step.
Causes
  • The application is unable to get the Lookup table entry for a catalog (for which you are doing prediction) from the following Lookup tables
    • Machine Learning Attributes Configuration Lookup
    • Machine Learning Services Threshold and Version Lookup (Categorization service only)
  • The application is unable to establish a MongoDB connection.
  • The application is unable to set Suggested Categories or Feedback attribute.
Solution
  • Check the workflow and then ML log of the service for which you have triggered training.
  • Check Machine Learning Attributes Configuration Lookup and Machine Learning Services Threshold and Version Lookup (Categorization service only) tables to see whether an entry for a catalog is correctly configured. For more information, see Using Machine learning (ML) assisted data stewardship.
  • Check whether the MongoDB details that are configured in the $TOP/mdmui/machinelearning/config/ml_configuration.ini file are correct and the same as configured in the env_settings.ini file.
  • Check whether a spec contains Suggested Categories and Feedback attributes. For more information, see Using Machine learning (ML) assisted data stewardship.