Amazon Forecast, launched in 2019, is a service that offers artificial intelligence (AI)-based forecasting via statistical and machine learning (ML) algorithms. However, many users have shifted their interests to Amazon SageMaker Canvas for benefits such as faster model building, lower costs, enhanced features and overall improved transparency.
SageMaker Canvas was launched as a response to the growing demands of Forecast users for upgraded AI capabilities. The platform offers an array of end-to-end ML solutions including, but not limited to, regression, multi-class classification, and natural language processing. With SageMaker Canvas, users can access these tools through a unified user-friendly platform.
On average, SageMaker Canvas provides a 50% faster model building performance and 45% quicker predictions for time-series models compared to Amazon Forecast. Also, SageMaker Canvas uses only the compute resources of Amazon SageMaker, making costs significantly more reasonable than Amazon Forecast. SageMaker Canvas not only allows direct access to trained models but also facilitates greater transparency by revealing details like hyperparameters used during training, model performance metrics, etc.
Further benefits come from Canvas’ ease of use, with features for automated missing data filling solutions and the ability to do what-if analysis directly using the interface. To respond to the growing user needs, Amazon will focus on updating SageMaker Canvas’ forecasting capabilities, including improved latency, reduced costs, accuracy enhancements, and better algorithms.
To assist users in transitioning from Forecast to SageMaker Canvas, Amazon has released a transition package including a workshop for hands-on experience and a Jupyter notebook to convert existing Forecast datasets into the SageMaker Canvas format.
Key differences between the two include the dataset types used, with Forecast requiring multiple types, whereas SageMaker Canvas uses only one. Additionally, SageMaker Canvas allows users to invoke models for a single or batch of datasets directly, unlike Forecast where a forecast needs to be created and then queried.
From the SageMaker Canvas UI, developers can join the target time series and other datasets into one. Also, they can build, test and deploy models via programmatic interactions utilizing SageMaker Canvas APIs. Furthermore, for increased flexibility, predictions can be generated and consumed in various ways, including in-app predictions and model deployment to a SageMaker endpoint.
In conclusion, the transition from Forecast to SageMaker Canvas promises enhanced user convenience, affordability, and greater transparency, making time series forecasting more intuitive and resource-effective.
