Settings tab on the Adaptive Model form

You can configure the update frequency and other settings that control how an adaptive model operates.

The settings in this tab are grouped into the following categories:

Standard settings

  • Update model after each n responses – Update model after n number of responses is received. When a model is updated, it is retrained with the specified number of responses and the resulting score is passed to each ADM client node and the Pega Platform components that are using the model.
  • Performance monitoring memory – The number of weighted responses used to calculate the model performance that is used in monitoring. The default setting is 0, which means that all historical data is to be used in performance monitoring.

Advanced data science settings

On update model:

  • Use all responses – For each update cycle, use all received responses.
  • Use subset of responses – Out of each update cycle, use only a specific number of randomly selected responses.

Data analysis binning:

  • Grouping granularity – A value between 0 and 1 that determines the granularity of the predictor binning. The higher the value, the more bins are created. The value represents a statistical threshold that indicates when predictor bins with similar behavior are merged. The default setting is 0.25.
  • Grouping minimum cases – A value between 0 and 1 that determines the minimum percentage of cases per interval. Higher values result in decreasing the number of groups, which can be used to increase the robustness of the model. Lower values result in increasing the number of groups, which can be used to increase the performance of the model. The default setting is 0.05.
    Note: This setting operates in conjunction with Grouping minimum cases to control how predictor grouping is established. The fact that a predictor has more groups typically increases the performance, however the model might become less robust.

Predictor selection:

  • Performance threshold – A value between 0 and 1 that determines the threshold for excluding poorly performing predictors. The default setting is 0.52.

    The value is measured as the coefficient of concordance (CoC) of the predictor as compared to the outcome. A higher value results in fewer predictors in the final model. The minimum performance of CoC is 0.5, therefore the value of the performance threshold should always be set to at least 0.5.

  • Correlation threshold – A value between 0 and 1 that determines the threshold for excluding correlated predictors. The default setting is 0.8. Predictors that have a mutual correlation above this threshold are considered similar, and only the best of those predictors are used for adaptive learning. The measure is the correlation between the probabilities of positive behavior of pairs of predictors.

Advanced technical settings

Attach audit notes to work object – Select this option if you want adaptive model details captured in the work object's history. This option is disabled by default.

Enabling this setting causes significant performance overhead.