Positron, the integrated development environment (IDE) by Posit for data science, now operates on Amazon SageMaker AI. This allows data teams to avoid working with separate tools for data access, analysis in R or Python, deployment, application development, and reporting. Using Positron on SageMaker AI provides advantages such as data access without credential management, as Positron operates under the Space execution role. This allows for queries to Amazon Athena, AWS Glue Data Catalog, and Amazon S3 directly from the IDE, following role permissions without the need to store or update keys. Computational power is also offered on demand. Users can run Space on the required instance size, and teams can reserve capacity using the SageMaker AI training plan for ongoing training. Additionally, the Posit Assistant, an AI assistant from Posit, can use Amazon Bedrock as a model provider, ensuring secure execution within the user's own AWS account.
The workflow demonstrates the use of a synthetic portfolio of 50,000 loans stored on Amazon S3 and registered in the AWS Glue Data Catalog. Amazon Athena executes queries on the data, R analysis checks features, and Python targets training an XGBoost classifier. The generated results are recorded and become part of the deployed application in Positron. The workflow also demonstrated that 50,000 loans were identified, of which 1,500 records lacked income data, and 1,015 default cases led to an overall default rate of 2.03 percent. The discussed results also showed that the AUC (area under the ROC curve) for the finally evaluated classifier was 0.834, with the highest risk among borrowers in the top decile being 12.3 percent.
The final results of the workflow include the deployment of the model, which became available via the InService channel. However, it should be noted that this workflow did not establish model fairness, calibration, lending suitability, production latency, or regulatory compliance. These points highlight that the conclusions drawn pertain only to a single session and do not serve as general performance or cost indicators. Thus, working with Positron in SageMaker AI demonstrated the integration capabilities of data science, analytical modeling, and application deployment in one space, providing a high degree of interaction and data accessibility.
Related articles
