Amazon SageMaker Canvas FAQs

SageMaker Canvas enables you to seamlessly discover AWS data sources that your account has access to including Amazon Simple Storage Service (S3), Amazon Athena (Glue Data Catalog), Amazon Redshift, Amazon Aurora, and Amazon RDS. SageMaker Canvas also supports external data sources including Salesforce Data Cloud, Snowflake, Databricks, and 40+ SaaS platforms. Finally, you can drag and drop files from your local disk to upload your dataset to SageMaker Canvas.

SageMaker Canvas supports importing tabular (CSV, Parquet), image (JPEG, PNG), and document data (PDF, JPG, PNG, TIFF).

You can analyze and explore your data in SageMaker Canvas using pre-built visualization or use natural language to generate custom visualizations. Amazon SageMaker Canvas also providers a Data Quality and Insight report to verify data quality (such as missing values, duplicate rows, and data types) and detect anomalies (such as outliers, class imbalance, and data leakage) in your data.

SageMaker Canvas offers a selection of over 300 prebuilt, PySpark-based data transformations, so you can transform your data and scale your data preparation workflow without writing a single line of code. Additionally, you can transform your data for ML models using FM-powered natural language instructions.

You can launch or schedule a job to quickly process your data or export the data preparation flow as a processing step in your ML workflow using integration with SageMaker Pipeline.

SageMaker Canvas provides a Data Quality and Insight report in the data preparation flow to check data quality and estimate model accuracy. It also validates your data prior to model building to check for common issues.

Yes, SageMaker Canvas provides access to ready-to-use foundation model (FMs) for content generation, text extraction, and text summarization. You can access FMs such as Claude 2, Llama-2, Amazon Titan, Jurassic-2, and Command (powered by Amazon Bedrock) as well as publicly available FMs such as Falcon, Flan-T5, Mistral, Dolly, and MPT (powered by SageMaker JumpStart) through a no-code interface and tune them using your own data.

SageMaker Canvas offers tabular, NLP, and for use cases including sentiment analysis, object detection in images, text detection in images, and entities extraction. These ready-to-use models do not require model building, and are powered by AWS AI services, including Amazon Rekognition, Amazon Textract, and Amazon Comprehend.

Currently, you can create classification (binary and multiple categories), regression, time-series forecasting, single label image classification, and multi-category text classification models in SageMaker Canvas.

SageMaker Canvas provides multiple options to build a model.

• Preview: This option lets you preview your model in about 2 minutes to give you an indicator of the model accuracy and feature importance.
• Quick Build: This option allows you to build a model quickly (approximately between 2 and 20 minutes) and provides a ready-made model.
• Standard Build: This option is extensive and may take a few hours depending on the size of your dataset. Standard build models provide you with detailed information including metric scores, training experiments using different combinations of hyperparameters, and generates multiple models in the backend. It then picks the best model that you can evaluate and use.

SageMaker Canvas provides column impact analysis which explains the impact of each column in your dataset on a model. SageMaker Canvas also provides additional metrics that provide visibility into model performance. Additionally, when you generate predictions, you can see the column impact that identifies which columns have the most impact on each prediction.

Yes. Data scientists can share any ML model built by other tools once it is registered in the SageMaker Model Registry, allowing you to generate predictions on these models in SageMaker Canvas.

To make a single prediction, go to the “single prediction” tab of the corresponding model version, input values, and SageMaker Canvas will show you the prediction. You can also use sliders and pull-down menus to change input values to see the impact on the prediction. To make predictions for multiple observations or rows of data, go to the “bulk prediction” tab, drag and drop the CSV, JPEG, or PNG file containing your observation, and SageMaker Canvas will create a new CSV, JPEG, or PNG file with predictions. SageMaker Canvas allows you to run manual and automatic batch predictions. Automatic batch prediction workflows get triggered every time an associated dataset is updated. You can then review prediction results inline or download for review.

You can select single or multiple batch predictions in SageMaker Canvas and send them to multiple Amazon QuickSight users in an account. You can open QuickSight with a single click from SageMaker Canvas, analyze the prediction as a dataset, build, and publish predictive dashboards that can be continuously updated for new and changed data.

Yes. You can share ML models built in SageMaker Canvas with data scientists working in SageMaker Studio. Data scientists can review, update, and share updated model versions with you, so you can generate predictions on the new versions in SageMaker Canvas.

After building and training a standard model in SageMaker Canvas, you can share your ML model using the share button in SageMaker Canvas. You can choose to share the model to a single user or multiple users within SageMaker Studio.

The ML model and artifacts shared from SageMaker Canvas will contain the dataset, data transformations (including the recipe data flow and transformation code), list of candidate models and the recommended model, data exploration report, candidate definition notebook, and explainability metrics (including feature importance).

Data scientists using SageMaker Studio can view model artifacts and recommend an alternate candidate from the list of candidates in SageMaker Autopilot. In addition, they can open and update data transformations with SageMaker Data Wrangler, update the model using SageMaker Autopilot, and share the new model version.

SageMaker Studio users can send updates on model versions using the share button within SageMaker Studio. The updated model from SageMaker Studio will appear as a new version of the original shared model directly in SageMaker Canvas.

Updated models are automatically versioned within SageMaker Canvas. You can access different versions of the model through the drop-down menu within SageMaker Canvas.

You can share standard build models containing tabular data for all use cases within SageMaker Canvas, including customer churn, predicting home prices, sales forecasting, predicting loan defaults, predicting hospital bed occupancy, and time-series forecasting models. You can also share custom CV and NLP models.

Yes. Once you create ML models in SageMaker Canvas, you can register them to SageMaker Model Registry and plug them into your existing model deployment CI/CD processes. SageMaker Model Registry is a repository that catalogs ML models, manages various model versions, associates metadata, manages the approval status of a model, and deploys them to production.

When you select a model version in SageMaker Canvas and register it to SageMaker Model Registry in your own account, SageMaker Canvas automatically sends model artifacts to SageMaker Model Registry such as reference links to the model inference container, model feature importance reports, model metadata such as training metrics, and associated charts. Once the model is registered, you can track the approval status in SageMaker Canvas. Rejecting a model in SageMaker Model Registry prevents the model from being deployed into an escalated environment, whereas approving a model in the SageMaker Model Registry can trigger a model promotion pipeline. The model promotion pipeline automatically copies the model to your pre-production AWS account, and triggers that the model is ready for inference workloads.

SageMaker Canvas offers ready-to-use foundation models (FMs) powered by Amazon Bedrock and SageMaker JumpStart. These models enable you to generate and summarize content. You can use natural language instructions to perform tasks such as creating narratives, reports, and blog posts; answering questions; summarizing notes and articles; and explaining concepts, without writing a single line of code. Your data is not used to improve the base models, is not shared with third-party model providers, and stays entirely within your secure AWS environment. With the same no-code interface, you can upload a dataset and select an FM, and SageMaker Canvas automatically helps you build custom foundation models to generate predictions immediately. SageMaker Canvas also displays performance metrics, so you can collaborate easily to generate predictions using FMs and understand how well the FM is performing on a given task.

SageMaker Canvas offers permissions for admins to control access to foundation models in the SageMaker Canvas user interface. This includes both features powered by foundation models such as data preparation and to ready-to-use foundation models.