The purpose of trainingMode="UPDATE"
is to process new items added to the items dataset (via PutItems or a bulk upload) as well as impression data for new interactions added to the interactions since the last FULL
/UPDATE
training. The new items and impressions are used to update the exploration feature for solutions created with the aws-user-personalization recipe. That is, to bring in new/cold items for exploration and to adjust probabilities for existing cold items in further exploration. Note that UPDATE
only brings in new items and impression data and does not retrain the model.
Therefore, if there are no dataset updates since the last FULL
/UPDATE
, there is no value in creating a new solution version with UPDATE
.
Finally, keep in mind that Personalize automatically updates solution versions created with the aws-user-personalization recipe every two hours at no cost. This essentially does an UPDATE
for you.
With User-Personalization, Amazon Personalize automatically updates the latest model (solution version) every two hours behind the scenes to include new data without creating a new solution version. With each update, Amazon Personalize updates the solution version with the latest item information and adjusts the exploration according to implicit feedback from users. This allows Amazon Personalize to gauge item quality based on new interactions for already explored items and continually update item exploration.
If you create a solution version with UPDATE
, you will be charged for the server hours to perform the update. Practically speaking, the only time you would need to manually create a solution version with UPDATE
is when you do not want to wait for the next automatic update.