Training Failed error when training a similarity model in Predictive Intelligence

boyan_kushlev
Kilo Explorer

I have a simple developer instance I requested from ServiceNow Developers a month ago. I want to use Predictive Intelligence to train a model on the existing knowledgebase articles in the database so that I can show related articles based on the current article's content. I installed the Predictive Intelligence plugin along with its dependencies.

When I go to Predictive Intelligence > Similarity > Solution Definitions and click on the Knowledge Similar Articles definition, I can see there are 1200 matching records from the kb_knowledge table. Fields are set to Short Description, Article Body, Wiki, and test fields - to Short Description. These are the default values. When I click on Update & Retrain at the bottom I can a training is scheduled in the Training Request Schedules tab. When I click on it and then click on Execute Now, and then go to the ML Solutions tab, I can see the state of the solution is showing "Configuration OR Network Error" (see attachment). I can keep retraining it, changing values, etc, etc, and all I get is this error. Once I retrain it, the previous solution would go to either "Training request timed out" or "Training is canceled".

I have no idea what is causing this issue. I found other people having a similar issue, but no resolutions whatsoever. Some respondents mentioned Predictive Intelligence doesn't work on personal developer instances, but this doesn't make a lot of sense - why is it on the plugin store in the first place? And if that's actually the case, what can I do to use it? Should I request a different type of instance? The documentation doesn't say anything about that.

11 REPLIES 11

Lener Pacania1
ServiceNow Employee
ServiceNow Employee

Hi Boyan, have you tried training any of the other out of the box models?  For example the out of the box classification model for assignment group?  

It says "Solution training requires a minimum number of 10000 records." So that's not really an option. The largest table I have is about 1300 records.

You can reset the default training limits by going into application navigator > sys_properties.list and changing the "glide.platform_ml.api.csv_min_line" value down to the 1k records (the default is 10k).

find_real_file.png

The 1,300 records are those incidents?

I get the exact same "Configuration OR Network Error". Can it be an issue with the data within the tables? Is there a demo table which this definitely should work for?