I'm using AWS Forecast managed service for predicting customer expenditure over time.
I understand I have to create a dataset group that contains the datasets: primaray, secondary and metadata. (And then import these into the group)
One thing I can't get my head around is how to organise my datasets, and if I need a "Dataset Group" for each customer, or if I can merge all customer expenditure data into the same dataset?
As I understand a Dataset Group can only contain a single primary dataset (merged customer data in this case). However, the spending by one customer has nothing to do with the spending of another. If merging all data into a single primary dataset, can I generate forecasts by filtering on the customer ID and these separate forecasts will not be related to each other in any way?.
Or, does this mean I have to create a Dataset Group per customer?
If merging all data into a single primary dataset, can I generate forecasts by filtering on the customer ID
Yes.
and these separate forecasts will not be related to each other in any way
You can query and forecast per item
AWS Forecast differentiates by three dataset types (target, related and meta) of which only target is mandatory. Speaking for DEMAND forecasting, the target dataset defines theses 3 attributes: date, item_id, demand. Further details on transactions will be stored in the related dataset.
I recommend theses samples: https://github.com/aws-samples/amazon-forecast-samples
And for detailed explanation check out the developer guide: https://docs.aws.amazon.com/forecast/latest/dg/forecast.dg.pdf