I am interested in the configuration of the adaptive decision manager settings in Pega 8.4:
Model and predictor data snapshot
Store all model data and only the latest predictor data
Store all model data and all predictor data
As of now the first setting is active, I am able to see the new graph points in the trend tab for each model report which are generated daily (as the snapshot schedule is based on an agent). Likewise, instances of classes Data-Decision-ADM-ModelSnapshot and Data-Decision-ADM-PredictorBinningSnapshot are also present for multiple days.
Is this setting above impacting the trend graph of the adaptive models' performance, if storing only the latest predictor data is selected?
If so, should the first option only allow to store the latest predictor data, delete any old data and always be represented as one point in the trend graph in the model report after each snapshot update? Would the previously gathered data be removed/ignored or invisible in the trend tab?
***Edited by Moderator: Pooja Gadige to add product details tag, add platform capability tag***
The default Adaptive Decision Manager settings saves all Model related data but only the latest Predictor data. When the ADM monitoring agent runs everyday, it will save the Model related information , delete earlier predictor related snapshot and save only the latest predictor snapshot in the ADM monitoring tables which are described in this link (https://community.pega.com/knowledgebase/articles/decision-management/database-tables-monitoring-models). If you would like a trend chart with even the predictor related snapshots then you will have to select the second option. However , this will lead to predictor related information being written to the predictor monitoring tables each time that the monitoring agent run and the table can quickly grow to have a large number of records depending on the number of predictors used in your models.
As Nithin points out, you can get Pega to keep predictor data over time but note there are no standard predictor-level trend charts. If you have the data you can build those plots using either Pega Reports or exporting the data and use other (BI) tools.
As an alternative option you could set up an ETL process to run daily copies of the predictor table from the datamart then use those as the basis for your trend charts.