Error Training Sentiment Analysis Model on Washington DC PDI (Prediction Error)

[jmariscalal]

Hello Community,

I'm trying to configure the Sentiment Analysis model for CSM Cases on my Washington DC PDI. I've followed the official documentation but I'm stuck at the model training stage.

I consistently receive the generic error: Your model didn’t finish training because of a prediction error. Review the training setup and launch it again.

I have gone through an exhaustive debugging process and wanted to share everything I've already tried to provide full context and avoid repeated suggestions:

Configuration and Troubleshooting Steps Performed:

1. Plugin Activation: The Task Intelligence for Customer Service (com.snc.csm_ml_task) plugin and all its dependencies are installed and active.

2. Test Data Creation: I created over 5 test cases to ensure the table was not empty. The text for analysis is correctly populated in the description field.

3. Isolated "Clean Slate" Test: To rule out data issues, I deleted all cases and ran a test with a single, "perfect" record. The error persisted.

4. System Property Check: I verified that the glide.shared_service_scheduler.url system property is active and has the correct value: https://sncmlscheduler.service-now.com/.

5. System User & Permissions: I confirmed that the sharedservice.worker system user is active and has the platform_ml role assigned.

6. Model Filters: I have verified that the model training is being launched without any filter conditions applied.

Log Analysis (The Key Evidence):

Finally, I checked the System Logs (sys_log) and found what appears to be the root cause—a connectivity issue:

Error DxC_ML: Failed to connect to scheduler at https://sncmlscheduler.service-now.com/... msg : com.snc.services.client.exception.ServerUnavailableException: Unable to connect to server...

Seeing this error in the logs, my conclusion is that this is a network-specific issue with my PDI that is preventing it from connecting to the ML services.

My question for the community is:

1. Is this a known issue with the current Washington DC PDIs?

2. Is there a workaround to complete the training on a PDI, or is this functionality effectively limited or disabled in these free developer environments?

I appreciate any information or confirmation you can provide.

Thank you!

[Dr Atul G- LNG · Accepted solution]

Hi @jmariscalal 

"As far as I know, these solutions don’t work on a PDI, as they require AI/ML features that aren't fully available in Personal Developer Instances. So in short, what you're trying to do may not work — or is likely not supported — in a PDI

[jmariscalal]

Thank you for your reply, I will try the implementation in my entreprise paid dev instances and let you know.

[Dr Atul G- LNG]

Sure mate. Could you please close the thread