Predictive intelligence is not working

[Smriti Rastogi]

I have a very simple use case of predicting the category on a change form on the basis of short_description.

It was working fine but suddenly I see no category being predicted.

Error - - Malformed URL.

Any quick help is appreciated.

[Smriti Rastogi]

The issue stems from the missing word vector artifact records in the [sys_attachment] table. The training is failing when it is trying to fetch these records from the instance and it returns a null, as these records do not exist.

There is a known issue with preserving solutions in a clone and this could be a cause for the missing word vector model artifacts that prevents the ML Solution training from completing successfully.

KB0861056: Clone does not preserve ML Solutions when setting property [glide.platform_ml.clone_artifacts] to true.

If you set the system property [glide.wordvector.upgrade_time_frame] from 180 [default] to 0, it will regenerate the word vector model each time you submit your ML Solution for training, and it will increase the time it takes to train the solution. However, it will regenerate all the Machine Learning (ML) artifacts, including the word vector artifacts that are missing that prevents it from being trained.

However we did not clone the system recently.

[Lener Pacania1]

Thanks for posting Smriti, I have seen that error when moving a PI model to a cloned environment or moving a PI model from one instance to another.  Glad to hear support was able to identify the issue and that your models are now training.

[Brian Bakker]

Hello all,

You will be pleased to know that the issue with the missing attachments on the [ml_model_artifact] records caused by excluding large attachments as a clone option has been resolved and these attachments are now transferred in the clone, even with the exclude large attachment option enabled.

Therefore, any instance cloned before 9th February 2022 will still have the issue of missing attachments and these [ml_model_artifact] records can be imported from an XML export of these records on the source clone instance.

Whereas, if the instance was cloned after this date, the attachments on the [ml_model_artifact] records will now exist on the target instance.

Best regards,

Brian