Database tables contain monitoring information for adaptive and predictive models. This data is used to populate the charts in the Monitor tab of every model rule in Dev Studio and additional reports in the Actions > Reports section of Prediction Studio.

The tables are transparently designed to contain monitoring data that you can easily use to create your custom reports in Pega Platform™ or any other reporting tool. Before you customize model reports to serve your business needs, learn what properties you can find in these tables.

For more information about how the Adaptive Decision Manager service captures and populates that monitoring data, see Configuring the Adaptive Decision Manager service.

Properties in the PR_DATA_DM_ADMMART_MDL_FACT table

Properties with the py prefix are defined on the Data-Decision-ADM-ModelSnapshot class. The out-of-the-box data set which applies to this class is pyModelSnapshots.

Properties in the PR_DATA_DM_ADMMART_PRED table

Properties with the py prefix are defined on the Data-Decision-ADM-PredictorBinningSnapshot class. The out-of-the-box data set which applies to this class is pyADMPredictorSnapshots.

Properties in the PR_DATA_DM_BINARY_DISTRIBUTION table (binary outcome predictive models)

For binary outcome models, the count of the positives and negatives is stored in a granular set of bins which are used to calculate the AUC and the ROC curves. The score distribution and the observed responses overlay occur during training.

Properties in the PR_DATA_DM_CONTINGENCYTABLE table (categorical outcome predictive models)

For categorical outcome models, the confusion matrix of responses is the main statistic. Each cell in the confusion matrix is stored as a record in the database table.

The confusion matrix is used to calculate the performance values, such as the F-statistic for the model or the accuracy for the classes.

Properties in the PR_DATA_DM_HISTOGRAM table (continuous outcome models)

For continuous outcome models, the difference between the predicted outcome and the actual outcome is used to measure the performance. The distribution of these residual values is stored in bins of equal interval size. The Information that is gathered in the bins is used to calculate the root-mean-square error (RMSE) and mean absolute error (MAE) performance statistics for the model.

Properties in the PR_DATA_DM_SNAPSHOTS table (snapshot summary)

The monitoring information that is stored in the monitoring data mart contains data that is related to the same point in time; that collection of monitoring data is called a snapshot. This table contains one record per snapshot per model.

This information can be linked to data in other tables that contain more detailed, binned information that is used to calculate the performance statistics.

One-time model data may consist of the following details:

• The number of responses in the snapshot
• The performance of the model during the snapshot
• The user that triggered the snapshot

Properties in the PR_DATA_DM_NOTIFICATION table

This table contains Prediction Studio notifications that inform about sudden drops in predictive performance, models with low performance, or other issues with models.