Member-only story
- Tip #1 Attempt Google’s walkthrough example before starting on your data. The reason why I say this is that I tried to do it with our data and failed. The import was successful with our data (after getting some errors which I solved), but the model was not recognising the data. The errors were logged but solutions were not commonly out there online. Therefore I suggest attempt Google’s walkthrough example with data that works to get an idea of what is needed! Note: I did not finish Google’s Walkthrough example because of Tip #2. The errors I received on our data and the solutions I implemented are in this article.
- Tip #2 Beware of the model training cost. I was unable to estimate this and the pricing estimation calculator does not feature the Recommendations AI service. I assume it will take trial and error. Google mentions the prices of node hours here, but node hours do not mean 1 actual hour, as they could scale the compute (which is cool but not cool to estimate costs!). I would love to hear the costs occurred by others. I could also be missing something! I was going to cut down the size of the Google walkthrough example from Tip #1, however, my time has run out and the Product Owner has moved this feature down the priority list.
I have posted 2 walkthroughs using the Recommendation AI service for those curious (one with Google’s example data and one our data) and here are my summary notes.
Happy recommending!
