Redshift ML / SageMaker - Deploy an existing model artifact to a Redshift Cluster

Question

Is it possible to deploy an existing model artifact from SageMaker to Redshift ML?

For example, with an **Aurora ML** you can reference a SageMaker endpoint and then use it as a UDF in a `SELECT` statement. 
**Redshift ML** works a bit differently - when you call `CREATE MODEL` - the model is trained with **SageMaker Autopilot** and then deployed to the **Redshift Cluster**.

What if I already have a trained model, can i deploy it to a Redshift Cluster and then use a UDF for Inference?

Accepted Answer

As of January 30 2021, you can't deploy an existing model artifact from SageMaker to Redshift ML directly with currently announced Redshift ML preview features. But you can reference  sagemaker endpoint through a lambda function and use that lambda function as an user defined function in Redshift.

Below would be the steps:

1. Train and deploy your SageMaker model in a SageMaker Endpoint. 
2. Use Lambda function to [reference sagemaker endpoint](https://aws.amazon.com/blogs/machine-learning/call-an-amazon-sagemaker-model-endpoint-using-amazon-api-gateway-and-aws-lambda/). 
3. Create a [Redshift Lambda UDF](https://aws.amazon.com/blogs/big-data/accessing-external-components-using-amazon-redshift-lambda-udfs/) referring above lambda function to run predictions.

Redshift ML / SageMaker - Deploy an existing model artifact to a Redshift Cluster

Contenido relevante