CODEWITHSUNDEEP
amazon-web-servicesamazon-redshiftamazon-forecast
Ankit Goel
Ankit Goel

Reputation: 360

AWS Forecasting from Large Time Series data stored on Redshift

I have sales data for last 3 years stored in Redshift and it updates daily. I want to start forecasting every week for next week (based on any algorithm as a start).

Since we have 10 million data point per day, I want to run the forecast directly in Redshift as a query and generate forecast from it.

What is the ideal method?

(Currently I am using HWES (Exponential Smoothening) and it is being run on a smaller data subset in pandas because of computational capacity constraints. Flexible to look at any forecasting model such as SARIMA, Deep AR if required)

Upvotes: 0

Views: 364

Answers (1)

Bill Weiner
Bill Weiner

Reputation: 11032

I don't think there is a way to run those processes directly on Redshift. I've used Redshift as a very fast tool for applying the insights from ML processes run on subset data onto the full data set. I had a process for finding the distances to multi-dimensional centroids where the centroids were determined by ML predictions. Insights from 10,000 points used analyzed in ML applied to 100's of millions of rows in Redshift. The trick is converting the math for applying the insights to SQL.

Upvotes: 0

Related Questions