Task Intelligence models need a lot of data to be effective, we recommend at least 10k records to train your Task Intelligence models . If you try to train your model with less than 10k records you will see the below error which prevents you from training your model.
There may be situations where you don't have 10k records or you are trying to prototype the model with a small dataset. This will be the first disclaimer, training with less than 10k records is not recommended and if you open a support case because the model does not behave as you expect support will immediately tell you to train with more data. Ok, with that out of the way here is how you change the defaults in the Task Intelligence Admin console to allow for smaller data sets to be used.
You will need MAINT access to lower the default minimum records to train the model by following the below steps. Open a support case to ask them to do steps 1 & 2 below.
- Go to All > sn_ti_admin_template.list this will open the Task Intelligence Admin Templates.
- Change the Minimum Number Records for the template IDs sn.csmpredictintel.csm.categorization and sn.csmpredictintel.csm.categorization.
3. Once support has lowered the default minimum records, Go back to All > Task Intelligence > Setup > Predict Case Fields > Set Up Model. Create your model.
Notice the 10k case warning that prevents you from training is no longer there. Click Launch Training.
4. You should be able to successfully train the model with the smaller data set.
Disclaimer number two, lowering the default minimum records to below 10k records can result in models with low accuracy. Do so at your own risk.
-Lener
Hi Lener,
Thank you for this post, that's very useful for exploring the functionality on demo instances.
While it does work on Utah, on Vancouver I don't see how the minimum number of records can be edited, as the corresponding entry is now read-only (see screenshot). Is there any other way to train Task Intelligence on Vancouver with less data?
Best,
Philipp
