Update the training variables
ABS provides an update.sh
script in the /pingidentity/abs/util
directory to update the training-related variables.
Using the script, you can update the following variables:
Name | Variable |
---|---|
Continuous learning |
|
Training period |
|
Threshold update period |
|
Window length |
|
You can update the training period when the system is already in a running state by using the script. Review the following use cases before changing the training and threshold period. In all the use cases, the default training period is assumed to be 24-hours.
If you want to extend the training period, it is a best practice to add new APIs after the training period is adjusted to avoid APIs completing a shorter training period. |
You can also use Global Configuration REST API to update the training variables. For more information see, Global configuration update REST API.
Update the training interval
-
Increase the training period: You can increase the training period by executing the update script.
-
Case 1 – The API model is under training, that is, the training period is not over.
System Behavior – In this case, if you increase the training period, for example, from 24 hours to 48 hours, the AI model trains based on the updated training period.
-
Case 2 – The API model has completed the training process.
System Behavior – Increasing the training period has no effect on trained APIs. Any new APIs will use the new training period.
-
-
Decrease the training period
You can decrease the training period by executing the update script.
-
Case 1 – The API model is in the training process but has not reached the duration of the new training period.
System Behavior – Decreasing the training period (for example, from 24 hours to 12 hours) shortens the training period to 12 hours for the APIs that have not completed the training process. If the API has completed 10 hours of training, then it will now complete its training period after 2 more hours.
-
Case 2 – The API model is in the training process and the new training duration is less than the current AI model trained duration.
System Behavior – In this case the API model stops training itself at the current time and moves to the prediction mode. For example, if the original training period was 24-hours and the AI model has been trained for 18-hours; at this time if the training period is reduced to 12-hours, the AI model stops training itself and moves to the prediction mode.
-
Case 3 – API model has completed the training process.
System Behavior – Decreasing the training period has no effect on trained APIs. Any new APIs will use the new training period.
-
Execute the update.sh
script
The update.sh
script is available in the /opt/pingidentity/abs/util
directory. Copy the script from the util
directory to your MongoDB primary node. The training period and threshold can be changed simultaneously or individually.
After executing the script, stop and start all ABS nodes for the updated values to take effect. |
Access script help by logging into the MongoDB primary machine and running the following command:
/opt/pingidentity/mongo/update.sh help
Example: Change the training period to 48 hours.
/opt/pingidentity/mongo/update.sh -u absuser -p abs123 --attack_initial_training 48 updating training_period to 48 WriteResult({ "nMatched" : 1, "nUpserted" : 0, "nModified" : 1 }) The current values of the variables are: attack_initial_training=48 attack_update_interval=24 api_discovery=false discovery_update_interval=1 continuous_learning=true discovery_initial_period=24 url_limit=100 response_size=100 window_length=24 discovery_subpath=3 percentage_diskusage_limit=80 Global Config successfully updated