I added the Customer fields as predictors in the Adaptive Model. The adaptive model configuration is like this:
In Interaction history, there already has more than 900 offer data that related to the outcome defined in Adaptive model. In monitoring tab, we can know that the model is running, but all predictors still inactive.
Q: How and when the predictors will be active?
***Edited by Moderator: Pooja Gadige to add capability tag***
Thank you for your reply, I think my predictors are still inactive because the AUC is below 0.52. And in the article, it said "add predictors that potentially have a higher correlation with the outcome". Here is another question:
1. What happens if we want to keep all customer fields as predictors? How long the AUC for the model will be updated to 0.52?
2. What happens if my predictors remain inactive? Would it impact the model and propensity?
3. If I change the AUC to 0.50, would it be fine? I know it is not the best practice, but my client want to have all customer fields as predictors. and will remove some predictors once the model learned
Posted: 4 months ago
Posted: 4 Mar 2022 2:35 EST
Nizam Mohammed (Nizam)
@Intandhp Do you know the volume of the responses sent to ADM? As per the settings, model update happens after every 200 responses. Just check whether the responses are passed to the ADM model or not. These are the potential reasons for this issue.
1) Predictors are going as null to ADM. (Can get this data by turning the historical data check box in ADM).
2) Symbolic predictors having more than two hundred distinct values. (Check the values which are being passed to ADM).
3) Load on ADM nodes while updating models. (Check the number responses and try to increase the frequency of the ADM model frequency based on the volume of responses.
4) Check the model update frequency time in the ADM settings.
In the Monitor tab of the adaptive model, can you see if any positive responses have been received?
In the Predictors section you can check if all of the predictors always have an AUC of exactly 50. Or what the maximum performance is for each predictor (is it always < 52?).
If that is working, you can also zoom into some of the Model reports. By clicking on selected predictors check if the values for those are fine (predictor values should not always have the same value, or always be missing)
In the monitor tab, I can only see if the action has a response. If I check further in the model report for related action, I can see there is a Positive response being recorded.
In the Predictors section, most of the predictor has a maximum performance of 50, but some of the predictors have maximum performance of 66.67, 75, and 100. But for the predictors that have a maximum performance of more than 50, the predictors are still inactive.