Amazon Forecast is a completely managed service that makes use of statistical and machine studying (ML) algorithms to ship extremely correct time sequence forecasts. Launched in August 2019, Forecast predates Amazon SageMaker Canvas, a preferred low-code no-code AWS instrument for constructing, customizing, and deploying ML fashions, together with time sequence forecasting fashions.
With SageMaker Canvas, you get faster model building, cost-effective predictions, superior options reminiscent of a mannequin leaderboard and algorithm choice, and enhanced transparency. You may as well both use the SageMaker Canvas UI, which offers a visible interface for constructing and deploying fashions while not having to put in writing any code or have any ML experience, or use its automated machine studying (AutoML) APIs for programmatic interactions.
On this publish, we offer an outline of the advantages SageMaker Canvas presents and particulars on how Forecast customers can transition their use instances to SageMaker Canvas.
Advantages of SageMaker Canvas
Forecast clients have been searching for larger transparency, decrease prices, quicker coaching, and enhanced controls for constructing time sequence ML fashions. In response to this suggestions, we have now made next-generation time sequence forecasting capabilities obtainable in SageMaker Canvas, which already presents a sturdy platform for making ready knowledge and constructing and deploying ML fashions. With the addition of forecasting, now you can entry end-to-end ML capabilities for a broad set of mannequin varieties—together with regression, multi-class classification, pc imaginative and prescient (CV), pure language processing (NLP), and generative synthetic intelligence (AI)—throughout the unified user-friendly platform of SageMaker Canvas.
SageMaker Canvas presents as much as 50% quicker mannequin constructing efficiency and as much as 45% faster predictions on common for time sequence fashions compared to Forecast throughout numerous benchmark datasets. Producing predictions is  considerably less expensive than Forecast, as a result of prices are primarily based solely on the Amazon SageMaker compute resources used. SageMaker Canvas additionally offers glorious mannequin transparency by providing direct entry to educated fashions, which you’ll deploy at your chosen location, together with quite a few mannequin perception experiences, together with entry to validation knowledge, model- and item-level efficiency metrics, and hyperparameters employed throughout coaching.
SageMaker Canvas contains the important thing capabilities present in Forecast, together with the flexibility to coach an ensemble of forecasting fashions utilizing each statistical and neural community algorithms. It creates one of the best mannequin to your dataset by producing base fashions for every algorithm, evaluating their efficiency, after which combining the top-performing fashions into an ensemble. This method leverages the strengths of various fashions to provide extra correct and strong forecasts. You have got the pliability to pick out one or a number of algorithms for mannequin creation, together with the aptitude to guage the affect of mannequin options on prediction accuracy. SageMaker Canvas simplifies your knowledge preparation with automated options for filling in lacking values, making your forecasting efforts as seamless as potential. It facilitates an out-of-the-box integration of exterior data, reminiscent of country-specific holidays, by way of easy UI choices or API configurations. You may as well reap the benefits of its data flow function to attach with exterior knowledge suppliers’ APIs to import knowledge, reminiscent of weather information. Moreover, you may conduct what-if analyses immediately within the SageMaker Canvas UI to discover how numerous eventualities would possibly have an effect on your outcomes.
We’ll proceed to innovate and ship cutting-edge, industry-leading forecasting capabilities by way of SageMaker Canvas by decreasing latency, decreasing coaching and prediction prices, and bettering accuracy. This contains increasing the vary of forecasting algorithms we assist and incorporating new superior algorithms to additional improve the mannequin constructing and prediction expertise.
Transitioning from Forecast to SageMaker Canvas
Right now, we’re releasing a transition bundle comprising two assets that will help you transition your utilization from Forecast to SageMaker Canvas. The primary part features a workshop to get hands-on expertise with the SageMaker Canvas UI and APIs and to learn to transition your utilization from Forecast to SageMaker Canvas. We additionally present a Jupyter pocket book that exhibits methods to remodel your present Forecast coaching datasets to the SageMaker Canvas format.
Earlier than we learn to construct forecast fashions in SageMaker Canvas utilizing your Forecast enter datasets, let’s perceive some key variations between Forecast and SageMaker Canvas:
- Dataset varieties – Forecast makes use of a number of datasets – goal time sequence, associated time sequence (non-obligatory), and merchandise metadata (non-obligatory). In distinction, SageMaker Canvas requires just one dataset, eliminating the necessity for managing a number of datasets.
- Mannequin invocation – SageMaker Canvas lets you invoke the mannequin for a single dataset or a batch of datasets utilizing the UI in addition to the APIs. In contrast to Forecast, which requires you to first create a forecast after which question it, you merely use the UI or API to invoke the endpoint the place the mannequin is deployed to generate forecasts. The SageMaker Canvas UI additionally provides you the choice to deploy the mannequin for inference on SageMaker real-time endpoints. With just some clicks, you may obtain an HTTPS endpoint that may be invoked from inside your utility to generate forecasts.
Within the following sections, we talk about the high-level steps for remodeling your knowledge, constructing a mannequin, and deploying a mannequin utilizing SageMaker Canvas utilizing both the UI or APIs.
Construct and deploy a mannequin utilizing the SageMaker Canvas UI
We advocate reorganizing your knowledge sources to immediately create a single dataset to be used with SageMaker Canvas. Confer with Time Series Forecasts in Amazon SageMaker Canvas  for steerage on structuring your enter dataset to construct a forecasting mannequin in SageMaker Canvas. Nonetheless, in the event you desire to proceed utilizing a number of datasets as you do in Forecast, you will have the next choices to merge them right into a single dataset supported by SageMaker Canvas:
- SageMaker Canvas UI – Use the SageMaker Canvas UI to affix the goal time sequence, associated time sequence, and merchandise metadata datasets into one dataset. The next screenshot exhibits an instance dataflow created in SageMaker Canvas to merge the three datasets into one SageMaker Canvas dataset.
- Python script – Use a Python script to merge the datasets. For pattern code and hands-on expertise in remodeling a number of Forecast datasets into one dataset for SageMaker Canvas, confer with this workshop.
When the dataset is prepared, use the SageMaker Canvas UI, obtainable on the SageMaker console, to load the dataset into the SageMaker Canvas utility, which makes use of AutoML to coach, construct, and deploy the mannequin for inference. The workshop exhibits methods to merge your datasets and construct the forecasting mannequin.
After the mannequin is constructed, there are a number of methods to generate and eat forecasts:
- Make an in-app prediction – You’ll be able to generate forecasts utilizing the SageMaker Canvas UI and export them to Amazon QuickSight utilizing built-in integration or obtain the prediction file to your native desktop. You may as well entry the generated predictions from the Amazon Simple Storage Service (Amazon S3) storage location the place SageMaker Canvas is configured to retailer mannequin artifacts, datasets, and different utility knowledge. Confer with Configure your Amazon S3 storage to study extra concerning the Amazon S3 storage location utilized by SageMaker Canvas.
- Deploy the mannequin to a SageMaker endpoint – You’ll be able to deploy the mannequin to SageMaker real-time endpoints immediately from the SageMaker Canvas UI. These endpoints will be queried by builders of their purposes with just a few strains of code. You’ll be able to replace the code in your present utility to invoke the deployed mannequin. Confer with the workshop for extra particulars.
Construct and deploy a mannequin utilizing the SageMaker Canvas (Autopilot) APIs
You should use the pattern code supplied within the notebook in the GitHub repo to course of your datasets, together with goal time sequence knowledge, associated time sequence knowledge, and merchandise metadata, right into a single dataset wanted by SageMaker Canvas APIs.
Subsequent, use the SageMaker AutoML API for time series forecasting to course of the information, prepare the ML mannequin, and deploy the mannequin programmatically. Confer with the pattern notebook in the GitHub repo for an in depth implementation on methods to prepare a time sequence mannequin and produce predictions utilizing the mannequin.
Confer with the workshop for extra hands-on expertise.
Conclusion
On this publish, we outlined steps to transition from Forecast and construct time sequence ML fashions in SageMaker Canvas, and supplied a knowledge transformation pocket book and prescriptive steerage by way of a workshop. After the transition, you may profit from a extra accessible UI, cost-effectiveness, and better transparency of the underlying AutoML API in SageMaker Canvas, democratizing time sequence forecasting inside your group and saving time and assets on mannequin coaching and deployment.
SageMaker Canvas will be accessed from the SageMaker console. Time sequence forecasting with Canvas is obtainable in all areas the place SageMaker Canvas is obtainable. For extra details about AWS Area availability, see AWS Services by Region.
Sources
For extra data, see the next assets:
Concerning the Authors
Dan Sinnreich is a Sr. Product Supervisor for Amazon SageMaker, centered on increasing no-code / low-code providers. He’s devoted to creating ML and generative AI extra accessible and making use of them to resolve difficult issues. Outdoors of labor, he will be discovered enjoying hockey, scuba diving, and studying science fiction.
Davide Gallitelli is a Specialist Options Architect for AI/ML within the EMEA area. He’s primarily based in Brussels and works carefully with buyer all through Benelux. He has been a developer since very younger, beginning to code on the age of seven. He began studying AI/ML in his later years of college, and has fallen in love with it since then.
Biswanath Hore is a Options Architect at Amazon Internet Companies. He works with clients early of their AWS journey, serving to them undertake cloud options to deal with their enterprise wants. He’s obsessed with Machine Studying and, outdoors of labor, loves spending time along with his household.
