Amazon is expanding its machine learning offering for business users with the new no-code, drag-and-drop service SageMaker Canvas. AWS also celebrated the success of SageMaker at re: Invent: Tens of thousands of customers would already use the machine learning workbench, which is primarily aimed at data scientists and ML developers.
Canvas is designed to allow the creation of ML predictions without further coding via a drag-and-drop interface. The application can be accessed via the SageMaker Studio console and the user interface can be set up in a few minutes of initial setup. The user can integrate ready-made CSV data sets from local data carriers or Amazon S3, if necessary also via Amazon Redshift or Snowflake.
Fast or Accurate?
When creating a model type, the application supports linear regression for forecasting, binary and multiclass logistic regression for classification, and time series forecasting. After implementing a data set, basic statistics and visualizations can be called up. Building a model in Quick Build mode provides quick, but less accurate results within a few minutes. Alternatively, the user can use the regular build process: The software works more precisely, but only outputs results after two to four hours.
Because Canvas model training is based on the same fundamentals as SageMaker AutoPilot and Amazon Forecast, the no-code alternative can deliver results that are as accurate as the code-based variants. Next to SageMaker Canvas AWS also presented the Secrets Detector for its CodeGuru at re: Invent, which developers should use to find passwords, API keys, SSH keys and access tokens in the source code and configuration files.