How does Adaptive Models learn (from implementation perspective) ?
Wanted to understand how the Adaptive Models learn. Based on the current implementation in the application I am working on, what I understand is that the Adaptive Models do not seek response data from PR_DATA_IH_FACT and related Dimension tables. As per the current implementation, it looks like the customer responses (embedded in offers) are also passed again through the same Adaptive Model. Is this the correct implementation approach?
Note: We are not using Run Interaction shapes in our application. Ours is a back-end service application with no UI.
**Moderation Team has archived post**
This post has been archived for educational purposes. Contents and links will no longer be updated. If you have the same/similar question, please write a new post.
Short answer is that ADM learns from associations between the predictor values and the outcomes. Typically (the "delayed learning" approach), the predictor values are stored at decision time and retrieved automatically by DSM when an outcome (or "response") is sent to ADM.
ADM and IH (to which the FACT table belongs that you refer), are unrelated; although some of the data stored in IH may be predictive.
What predictors to use is up to the implementation. Typically, this will be a mix of customer attributes, behavioral data and other context that might be relevant to predict the behavior of interest.
I hope this and the above PDN link provided by WaikeiKwok helps. You may want to consult a decisioning architect on the team or seek further answers in the academy courses available from Pega. Feel free to post specific questions to which you cannot find an answer on PDN of course.
In the delay learning adaptive models, you don't need to pass the response from the same adaptive models again. Pega stores the ADM inputs into a property called pxADMInputs with values when the request is sent. It uses this JSON data to retrain the models once the response is received. It wont capture the response from the IH table but it uses DDS to map the response which is an internal cassandra store.