- 최신
- 최다 투표
- 가장 많은 댓글
Hello @MK, thanks for re:posting this question, and glad to see that you're starting your ML journey with SageMaker Canvas! :D
As you have had the possibility to test and discover so far, Canvas is meant to be a UI-driven ML workflow enabler for business analysts and developers, with little to no knowledge about ML. Therefore, at this stage, we do not provide an API or an SDK to connect and use Canvas capabilities. Sorry about that! This situation might change in the future, so stay tuned.
Since that's the situation, I suggest you keep using Canvas for your "Exploratory data analysis" and your "Model experimentation" via Preview Models and Quick builds. I can tell you how you could automate some of these capabilities externally from SageMaker Canvas, with two capabilities:
-
Create and automate your ETL workflow with AWS Glue DataBrew: AWS Glue DataBrew provides a UI interface for tabular data analysis and transformation, plus it supports a plethora of sources including JDBC connector to MySQL servers! Once you have defined your transformation (maybe even quality rules) you can create a Recipe job that can be scheduled to your preference, so that you have your ETL process automated. Learn how to automate/schedule Glue Databrew in this blog post.
-
Train (and deploy) your model with SageMaker AutoPilot: SageMaker AutoPilot uses the same AutoML technology as SageMaker Canvas. This means that all you need to do is just to point to a specific S3 bucket where your CSV file is stored, and then call an API to get a model trained! You can set-up a Lambda function / StepFunction workload that triggers on a new file generated in the S3 bucket output of the recipe job above, which triggers an AutoPilot job. The output will be available either in SageMaker Studio or you can also describe the AutoML job to get the best candidate and deploy it using the information in
InferenceContainers
and calling, in order: -
create_endpoint_config() - a suggestion: use Serverless endpoint to save money!
Hope this helps!