- Newest
- Most votes
- Most comments
Hey CarcharodonCarcharias,
As far as I understand, you must meet the 1000 minimum requirement to train a solution version. You cannot get a working solution version/campaign without it. Once you have the model trained on the 1000 interactions, you can incrementally add to this by using putEvents, putItems, and putUsers, but you will not see the changes from these in the getRecommendations (at least for the putevents actions) until you have successfully retrained a new solution version.
Workaround: Since you may have the same problem as me, and simply do not have 1000 interactions available, you can try making some "dummy data" to get the model going, set all the dummy data with a "DUMMY" tag in the ITEMS.csv as metadata, and then remove all of these when getting recommendations using a filter. That seems to work well for me.
I hope that helps, let me know if there's any confusion.
Best,
Lawrence
Hello Lawrence
Thanks for sharing your insights. Based on your experience it does seem like I cannot build a corpus of 1000_ events via the putEvents API and then train a model. I MUST import a csv file with 1000_ interactions and train a model on it and then add subsequent events via the putEvents API.
This helped to clear a few things.
Thanks
Relevant content
- asked 3 months ago
- asked 6 months ago
- AWS OFFICIALUpdated a year ago
- AWS OFFICIALUpdated a year ago
- AWS OFFICIALUpdated 8 months ago
- AWS OFFICIALUpdated a year ago