improving-forecast-accuracy-with-machine-learning
improving-forecast-accuracy-with-machine-learning copied to clipboard
Deploy with data downloader not automatically kicking off pipeline
In the past I've mainly used the Amazon Forecast pre-PoC workshop templates to deploy this solution with demo data, but recently thought since the upgrade came out I'd try the direct method from the canonical template listed on the solution page.
Unfortunately, seems like (because of the dependency order of resources) the standard install template doesn't wait for state machine triggers to be configured before loading the demo data to the data bucket... And therefore doesn't automatically kick off the first round of forecasting.
For me this is not ideal - it's helpful for the one-click deploy with demo data to also trigger the Forecast pipeline for easy demonstration of the functionality. At the moment it seems we need to manually kick the pipeline by e.g. downloading and re-uploading one of the S3 data files.
Good catch -
The change required is setting a dependency order in the CDK code to ensure the nested "Demo" stack is only deployed after the S3 event notification has completed creation.