Build predictive models with SageMaker Canvas
QuickSight authors can export data into SageMaker Canvas to build ML models that can be sent back to QuickSight. Authors can use these ML models to augment their datasets with predictive analytics that can be used to build analyses and dashboards.
Prerequisites
-
A QuickSight account that's integrated with IAM Identity Center. If your QuickSight account isn't integrated with IAM Identity Center, create a new QuickSight account and choose Use IAM Identity Center enabled application as the identity provider.
-
For more information on IAM Identity Center, see Getting started.
-
To learn more about integrating your QuickSight with IAM Identity Center, see Configure your Amazon QuickSight account with IAM Identity Center.
-
To import assets from an existing QuickSight account to a new QuickSight account that's integrated with IAM Identity Center, see Asset bundle operations.
-
-
A new SageMaker domain that is integrated with IAM Identity Center. For more information about onboarding to SageMaker Domain with IAM Identity Center, see Onboard to SageMaker Domain using IAM Identity Center.
Topics
Build a predictive model in SageMaker Canvas from Amazon QuickSight
To build a predictive model in SageMaker Canvas
-
Log in to QuickSight and navigate to the tabular table or pivot table that you want to create a predictive model for.
-
Open the on-visual menu and choose Build a predictive model.
-
In the Build a predictive model in SageMaker Canvas pop up that appears, review the information presented and then choose EXPORT DATA TO SAGEMAKER CANVAS.
-
In the Exports pane that appears, choose GO TO SAGEMAKER CANVAS when the export is completed to go to the SageMaker Canvas console.
-
In SageMaker Canvas, create a predictive model with the data that you exported from QuickSight. You can choose to follow a guided tour that helps you create the predictive model, or you can skip the tour and work at your own pace. For more information about creating a predictive model in SageMaker Canvas, see Build a model.
-
Send the predictive model back to QuickSight. For more information about sending a model from SageMaker Canvas to Amazon QuickSight, see Send your model to Amazon QuickSight.
Create a dataset with a SageMaker Canvas model
After you create a predictive model in SageMaker Canvas and send it back to QuickSight, use the new model to create a new dataset or apply it to an existing dataset.
To add a predictive field to a dataset
-
Open the QuickSight console, navigate to the Datasets page, and choose Datasets.
-
Upload a new dataset or choose an existing dataset.
-
Choose Edit.
-
On the dataset' data prep page, choose ADD, and then choose Add predictive field to open the Augment with SageMaker modal.
-
For Model, choose the model that you sent to QuickSight from SageMaker Canvas. The schema file automatically populates in the Advanced settings pane. Review the inputs, and then choose Next.
-
On the Review outputs pane, enter a field name and description for a colum to be targeted by the model that you created in SageMaker Canvas.
-
When you are finished, choose Prepare data.
-
After you choose Prepare data, you are redirected to the dataset page. To publish the new dataset, choose, Publish & Visuallize.
When you publish a new dataset that uses a model from SageMaker Canvas, the data is imported into SPICE and a batch inference job begins in SageMaker. It can take up to 10 minutes for these processes to complete.
Considerations
The following limitations apply to the creation of SageMaker Canvas models with QuickSight data.
-
The Build a predictive model option that is used to send data to SageMaker Canvas is only available on table and tabular pivot table visuals. The table or pivot table visual must have between 2 and 1,000 fields and at least 500 rows.
-
Datasets that contain integer or geographic data types will experience schema mapping errors when you add a predictive field to the dataset. To resolve this issue, remove the integer or geographic data types from the dataset or convert them to a new data type.