aws personalize Maximum number of items that are considered by a model during training

Question

https://docs.aws.amazon.com/personalize/latest/dg/limits.html

says

Maximum number of items that are considered by a model during training

I have fare more items than 750k. Is there any workaround for this limitation?

Answer 1

This is currently a hard limit of the service. Some things to keep in mind.

• Training will still complete successfully if you have > 750k items.
• This limit applies to items that have interactions. In other words, the distinct number of items in your interactions dataset. These are warm items.
• Items that are in your items dataset but not in the interactions dataset are cold items. For the user-personalization, recommended-for-you (e-comm), and top-picks-for-you (VOD) recipes, Personalize will pull cold items from these uninteracted items for exploration. The total number of warm and cold items considered by the model is still limited to 750k.
• The cold items are updated automatically every two hours in the background. This is not a retraining, though. It just updates exploration. Again, just applies to the user-personalization, recommended-for-you (e-comm), and top-picks-for-you (VOD)recipes.
• When you retrain a model (trainingMode=FULL), the warm items are selected again from your interactions dataset based on engagement from users. This applies to all custom recipes. The recipes in domain dataset groups for e-comm and VOD do automatic full retraining weekly.

Depending on your use cases and how your users interact with items, Personalize can still be used successfully with more than 750k items.