Hello All,

1) According to the formula for SmoothedPropensity it is only  StartingEvidence  that matters - ((.StartingPropensity * .StartingEvidence) + (.pyPropensity * .ADMEvidence))/(.StartingEvidence + .ADMEvidence) since everywhere around ADMEvidence is not defined and is only referenced in strategies. @Guru, could you please clarify the meaning of ADMEvidence and why it is not defined? (Out of definition we still need a pair of Starting value for propensity and evidence defined, see PDMA Student Guide)

2) For my state of the VM I discovered the following offers that have no value for StartingEvidence

handsets/iphoneoffer
handsets/iphoneupgrade

plans/simpleoffer

plans/generousoffer

3) These seem to have been shipped with exercises or created manually there though with no hint to specify values for StartingEvidence and or Starting Propensity... which are not mandatory both technically and from their semantics.

Thanks for clarification,

Andrey 

Andrey Thank you for brining this to the attention of the community. Let me start by giving some background. 'Smooth Propensity' is a technique used to "smooth" the erratic output from Adaptive Models during initial days. Here's an excerpt from the 'Predicting Customer Behavior Using Real-time Data (7.1)' course (short code: CBRTD) ##################################### Smooth Propensity ----------------- Any strategy pattern which you implement should take into account that in the early days of the model the predictions are likely to be erratic and as the model matures, the predictions get better. To help us to use the model at the beginning we have introduced a concept of Smooth Propensity. We use a propensity smoothing mechanism to 'jump start' the process. Propensity smoothing requires a starting (assumed) propensity and starting evidence (number of assumed proposition responses that contributed to that propensity to be set for each proposition). When there are no responses, the smoothed propensity equals the starting propensity. The example graph and the table below shows how volatile is the model propensity following the initial responses to an offer. ##################################### Note: Please go through the above mentioned course for more details. ##################################### The formula for calculating 'Smooth Propensity' is ((.StartingPropensity * .StartingEvidence) + (.pyPropensity * .ADMEvidence))/(.StartingEvidence + .ADMEvidence) The out-of-the-box Next-Best-Action strategies used by the Next-Best-Action Advisor uses this concept. Hence all out-of-the-box propositions have the constituent attributes (Starting Propensity and Starting Evidence) populated with values. pyPropensity and ADMEvidence come from the Adaptive Models used in the Next-Best-Action Advisor strategies. Offers shipped in most of the 'Advanced' courses have the Starting Propensity and Starting Evidence attributes populated. If you find any Offers (either created by you or shipped in any exercise JARs) with these attributes not populated AND you hit the error being discussed when using NBAA, please set these values. Hope this clarifies. Regards Guru