Updating of Adaptive Model
Hi all,
Is my conclusion justified that the only way to run and update an adaptive model on a case to case base is by calling a data flow via an utility in the process flow?
My current implementation is like this:
- A process flow for a single case type
- In this process flow I've included an utility shape to run an activity with method: Call Data-Decision-DDF-RunOptions.pxRunSingleCaseDDF. In this dataflow my "determine recommendation" decision strategy is included
- After some other actions in my UI I've included another utility shape to run an activity with the same method, leading to my "capture results" decision strategy.
With this implementation both my Interaction History and Adaptive Model get updated, but it feels a bit devious to do it like this via data flows. Is there a way to achieve the same results by using interaction rules?
Thanks in advance!
Accepted Solution
Pegasystems Inc.
NL
Hi Gerli,
You are right in pointing out that disconnect. To clarify the position of the examples in DMSample, the reason why the sample application took its time to implement this new pattern was exactly because the integration with process flows lagged behind. Ultimately, we had to go the route of data flows (hence utility shape) because we needed to leverage new functionality. The good news is that, since v7.3.1, we no longer need activities when making decisions and capturing responses in case management–the process flow has now a dedicated shape for this. I do not know, however, if upgrading is a possibility in your case.
Interaction rules are supported by an implicit chain of triggers, which include adaptive learning; if adaptive models are in the decision path, the interaction rule automatically sends results to adaptive decision manager. One of the key things to understand is that 1) interaction rules served just one purpose, allowing for very little control beyond the implicit automatisms, but making sure that the right operations happened in the background whilst 2) data flows are powerful rules that serve a broad set of use cases, allowing for total control over the data pipeline–the consequence, of course, is that they require us to explicitly set destinations.
Does this help?
Regards,
Iolanda
Pegasystems Inc.
NL
Hi
You can still use interaction rules through the "Run Interaction" shape or via an activity. Although not explicit in the rule form, they can invoke Adaptive. This is documented on PDN, e.g. https://community.pega.com/sites/default/files/help_v731/procomhelpmain.htm.
However the preferred way is to use a data flow. With this approach you are flexible in where to send the strategy results: to IH, to the customer movie, to Adaptive, to data sets of your own, etc. The documented patterns for headless decisioning might help, see e.g. https://docs-previous.pega.com/decision-management-reference-materials/headless-decisioning-beginning-pega-722.
Regards
Otto
Atos Consulting
NL
Hi Otto,
I don't understand your answer completely, can you be a bit more specific?
You wrote that I can use the Run Interaction shape, or an activity, but the preferred way is to use a data flow. How can I call my data flow without using an activity? My LSA forbids me to use an activity, therefor I'm trying to find other ways to implement it.
I've read the documentation you referred to regarding the implicit invoke of Adaptive in "Run Interaction" shapes. In the documentation I can't find how to let the Run interaction shape update the adaptive model. I'm able to update the Interaction History by using the "Run Interaction" shape, but then my adaptive model doesn't count the responses. Can you explain in detailed steps how to do this?
Thanks!
Regards,
Gerli
Pegasystems Inc.
NL
Hi Gerli,
The documentation mentioned by Otto describes the preferred pattern, it uses a service activity for headless decisioning.
- PDN Tutorial: https://docs-previous.pega.com/decision-management-reference-materials/tutorial-creating-decision-strategies-headless-decisioning-scenario
- PDN Tutorial: https://pdn.pega.com/tutorial-building-headless-decisioning-scenario-data-flows
You will find an example of this in DMSample.
I would also highly recommend the 'Implementing Adaptive Decisioning' module on Pega academy:
https://academy.pega.com/search
Adaptive Models learn from responses flowed to pxAdaptiveAnalytics (not from Interaction History).
BTW which version of the Pega platform are you using?
Ivar
Atos Consulting
NL
Hi Ivar,
Thanks for your response.
I'm using the Pega 7.2.1 version.
Currently my implementation is like described in the "Implementing Adaptive Decisioning" module, which is by using data flows and in there my responses flow to pxAdaptiveAnalytics. I've found the ServiceActivity of the headless decisioning pattern but if I'm right this one is not referenced out of a process flow but stand-alone.
I took a look in the DMSample you are referring to and I see that all decision strategies in there are started via an interaction rule in the process flow. How does .pxAdaptiveAnalytics gets updated in those use cases? Is it because the decision strategy results are written to Visual Business Director?
To be honest, I'm a bit confused by the documentation and examples. All documentation seems to say that I should use data flows but the example just uses interaction rules. In my case I want to update the adaptive model with responses that are captured during single cases. To be clear, my current implementation is working fine but since I'm using activities to start the data flows my LSA asked me to find another solution in which we can get rid of the activities.
Regards Gerli
Accepted Solution
Pegasystems Inc.
NL
Hi Gerli,
You are right in pointing out that disconnect. To clarify the position of the examples in DMSample, the reason why the sample application took its time to implement this new pattern was exactly because the integration with process flows lagged behind. Ultimately, we had to go the route of data flows (hence utility shape) because we needed to leverage new functionality. The good news is that, since v7.3.1, we no longer need activities when making decisions and capturing responses in case management–the process flow has now a dedicated shape for this. I do not know, however, if upgrading is a possibility in your case.
Interaction rules are supported by an implicit chain of triggers, which include adaptive learning; if adaptive models are in the decision path, the interaction rule automatically sends results to adaptive decision manager. One of the key things to understand is that 1) interaction rules served just one purpose, allowing for very little control beyond the implicit automatisms, but making sure that the right operations happened in the background whilst 2) data flows are powerful rules that serve a broad set of use cases, allowing for total control over the data pipeline–the consequence, of course, is that they require us to explicitly set destinations.
Does this help?
Regards,
Iolanda
Atos Consulting
NL
Hi Iolanda,
Thanks for this extensive reply.
I've tried with the interaction rules to send its results to Adaptive Decision Manager, this didn't work unfortunately. But I understand from your response that interaction rules aren't the sustainable way to organize adaptive learning so I will not investigate on this further and go with the data flows.
Once it's possible to update to v7.3.1 we can move to the dedicated shapes with data flows and take out the activities again.
Regards Gerli