Adaptive decision manager services alerts
Hi,
I wanted to know what kind of alerts are being generated by adaptive decision manager services?i.e. does the data analysis which is done by the ADM services generate any alert ? What are those and when are they generated.
Thanks in advance,
Aniket
***Edited by Moderator Marissa to update platform capability tags****
Pegasystems Inc.
US
check this link: https://community.pega.com/knowledgebase/performance-and-security-alerts-pega-platform
There are several ADM related alerts (e.g., Pega0065, Pega0070).
Cognizant Technology Solutions
IN
Thanks for responding. One question that I have is where does the data analysis take place? I mean does this data analysis and refreshing of models have any adverse effect on the database and on Pega interaction ? Can this lead to any PEGA0005 alerts (i.e. while querying Interaction History for fetching customer interactions) ?
I am of the idea that the ADM services are running on dedicated ADM server and should not have any adverse effect otherwise. Is that understanding correct?
Pegasystems Inc.
US
Yes or no. Yes, ADM should be run on its own nodes (especially in production env) but the IH is all stored in the same table. A properly sized/tuned env should not have PEGA0005 alerts but it can happen for an environment, especially IH table is getting big. You should perform archiving on those IH tables to avoid potential performance issue.
Cognizant Technology Solutions
IN
Yeah Exactly. That's what I wanted to confirm that ADM data analysis efforts alone should not cause any PEGA0005 on IH tables or overhead on IH tables as it should happen asynchronously/remotely and not during decisioning. Right?
Cognizant Technology Solutions
IN
Also it will be really useful if someone can explain whether and how adaptive models hamper the performance ?
Does adaptive models take up records in Interaction History for calculating propensity?
When the ADM services refresh /update the adaptive models, does it have any adverse effect on the overall database and thus on performance?
Thanks.
Pegasystems Inc.
NL
Hi ANIKETD5,
The updating (learning/training) of adaptive models is decoupled from the scoring of the models and therefore doesn't have an adverse effect (as mentioned above, in a production environment there would be dedicated ADM worker nodes for updating the models).
When updating a model, the Interaction History database doesn't play any role.
Of course there may be valuable predictive information in the Interaction History. Usage of IH data as predictors is optional. When any strategy is executed that retrieves IH info (for ADM or other purposes) this can result in a database query on IH and as such be part of the strategy execution performance.
In 7.4, an alternative way to retrieve IH data in strategies was introduced that doesn't make use of a query, for better performance. IH summaries eagerly aggregate data from IH (when they are 'materialized').
https://community.pega.com/knowledgebase/release-note/aggregate-data-interaction-history-summaries
Consequently, in 8.1 we introduced an easy way to use these IH summaries as predictors to adaptive models, see:
Hi ANIKETD5,
The updating (learning/training) of adaptive models is decoupled from the scoring of the models and therefore doesn't have an adverse effect (as mentioned above, in a production environment there would be dedicated ADM worker nodes for updating the models).
When updating a model, the Interaction History database doesn't play any role.
Of course there may be valuable predictive information in the Interaction History. Usage of IH data as predictors is optional. When any strategy is executed that retrieves IH info (for ADM or other purposes) this can result in a database query on IH and as such be part of the strategy execution performance.
In 7.4, an alternative way to retrieve IH data in strategies was introduced that doesn't make use of a query, for better performance. IH summaries eagerly aggregate data from IH (when they are 'materialized').
https://community.pega.com/knowledgebase/release-note/aggregate-data-interaction-history-summaries
Consequently, in 8.1 we introduced an easy way to use these IH summaries as predictors to adaptive models, see:
https://community.pega.com/knowledgebase/articles/add-predictors-based-interaction-history
Regards,
Ivar
P.S. For more background on how ADM works, see also https://academy.pega.com/library/72/implementing-adaptive-decisioning