Posted On: Sep 28, 2023
Amazon SageMaker Canvas now offers a quicker and more user-friendly way to create machine learning (ML) models for time-series forecasting. With its visual, point-and-click interface, business analysts can easily create accurate ML models for insights and predictions, without writing any code or having prior ML knowledge. Canvas supports multiple use cases — including time-series forecasting used in sectors such as retail, manufacturing, and finance — by combining statistical and ML algorithms to generate highly accurate forecasts.
Today, Canvas introduced upgrades to its forecasting capabilities to improve accuracy, enable faster model training and predictions, and support programmatic access. Compared to previous versions, you can now train a forecasting model up to 50% faster across various benchmark datasets, saving on average 110 minutes for data batches up to 100 MB. Generating predictions is also up to 45% faster, cutting the prediction time by an average of 15 minutes for a typical batch of 750 time series. Additionally, you can now regenerate predictions from an existing model by simply adding recent data, without having to retrain the model.
You can now programmatically access the model building and prediction functions via APIs, including comprehensive model accuracy and performance reports. These reports allow you to better understand how dataset attributes affect specific forecasts and gain deeper insights into the optimal models AutoML selects.
The improved forecasting is now available in all AWS regions where Canvas is supported. SageMaker instance charges will apply to generate forecast predictions.. To learn more, see the documentation and the Canvas pricing.