Amazon Forecast - extra attributes in the TARGET_TIME_SERIES?

Question

On Amazon Forecast, let’s say I have a variable (Z) which is supposed to help me predict a set of target variables (Y1, Y2, Y3).

First question is, what is the difference between:

1. put Z as an extra attribute in the TARGET_TIME_SERIES, that is, as an extra column
2. put Z as an attribute in the RELATED_TIME_SERIES

Second question is, given that Z has just one value per day (let’s say this is a stock price), how should I deal with the fact that I have 3x-repeated timestamps? Should I just repeat Z for each repeated date?

I understand that, if I'm not training my model to predict Z, I need to provide future values for it. But this makes option 1) even it more confusing to me. In which cases should one add an extra attribute in TARGET_TIME_SERIES?

Answer 1

I got a very nice explanation of it in here, so I'll leave it as the answer: https://github.com/aws-samples/amazon-forecast-samples/issues/104#issuecomment-764896502