Prediction Model Retrain

[Supriya39]

Hi All,

We have prediction model to predict Assignment Group and Archive Reason on call data. These models get called in Flow which set the values of Call's Assignment Group and Archive Reason. These models get trained every week.

Sometimes models does not get trained properly. The errors received are as below :-

404: Prediction Failed with error: TypeError: Expected argument of type object, but instead had type undefined. Please contact Support for Additional Assistance

404: Prediction Failed with error: java.lang.Exception: java.lang.Exception: 400 : Read timed out. Please contact Support for Additional Assistance

As a result the predictions are failed and Calls remain unarchived or unassigned. Unfortunately when we check this manually, we come to know about issue and then we retrain the model.

I need some suggestions on how to retrain these models upon receiving certain prediction failure through FLOW / any other way. 

Any help is appreciated. 

Thanks in advance. 🙂

[Amit Verma]

Hi @Supriya39 

Aren't you receiving any emails OOB from the ServiceNow instance whenever the training fails ? If not, we could approach this problem by two ways :

1. Create an After Update Business rule on the table "ml_solution" with trigger condition as State changes is Error While training solution. Inside the business rule, you could make use of Machine Learning APIs to retrain the model. Refer https://docs.servicenow.com/bundle/tokyo-now-intelligence/page/administer/predictive-intelligence/co...

https://docs.servicenow.com/bundle/tokyo-api-reference/page/app-store/dev_portal/API_reference/Class...

2. Create a flow with trigger as Record Updated on the table "ml_solution" and condition as State Changes to Error While Training Solution. Inside the flow, use the machine learning APIs to restart the training. 

Thanks & Regards

Amit Verma