How to create a continuous flow of forecast?

0

Hello: After playing a little bit with AWS Forecast, and starting seeing some good results, I would like to understand the Full Forecasting Flow, to see how I could put this into production... It seems that the flow starts once we accumulate enough data to start training the predictor, which can be trained for let's say 7 days, and once the model is ready, I can create the forecast for the next 7 days and consume this. But what about the following 7 days? I see the model does not allow to create forecast for these following 7 days.... so what are the next steps? Do we need a new model, or re-train the current one? Do we need to feed the real data from the last week? I'm not clear about this and the best way to create a continuous flow of forecast. I would appreciate any comment on this. Thanks

已提問 1 年前檢視次數 258 次
1 個回答
0

Hi User,

You will need to set up an automated workflow to continuously ingest the latest data and make predictions with the new data. As this is a time series problem, it can only make predictions with the most recent x timeframe of data. Here is an Architecture that suggests how such a workflow can be setup on AWS

Enter image description here

Link to the blogpost: https://docs.aws.amazon.com/forecast/latest/dg/tutorial-cloudformation.html

Minimally, you should have the following setup:

  1. Publishing of new data into S3
  2. Lambda function that get triggers when new object is put into S3, make predictions with the new data, and send it to SNS
  3. Your applications can subscribe to the SNS to receive the next time period's prediction

You can enhance your workflow to continuously train and evaluate your model, or integrate with other services for analysis etc as shown in the diagram.

Hope this information is helpful in your exploration of setting up an automated forecasting workflow on AWS!

profile pictureAWS
專家
ljunkai
已回答 1 年前
  • Thanks, junkai. The link actually points to this post, not to the article you mentioned.

  • I have fixed the link, you may try again. thanks!

您尚未登入。 登入 去張貼答案。

一個好的回答可以清楚地回答問題並提供建設性的意見回饋,同時有助於提問者的專業成長。

回答問題指南