How to make a SageMaker endpoint publicly available?

[stefanlinner]

Hello,

First of all, thank you so much for your packages, they really help me a lot!
This is not an issue, but a question I have.

I managed to deploy my machine learning model fully programatically using vetiver::vetiver_sm_build(), vetiver::vetiver_sm_model() and vetiver::vetiver_sm_endpoint() to Amazon SageMaker. I can also make predictions using the created endpoint with predict(new_endpoint, new_data). However, this only works if my AWS credentials are available in the project. My goal is to expose the model as a public API, i.e. anyone who knows the link can call it without needing my AWS credentials.

After doing a bit of research, I found the following post, which tells me that I need to use Amazon API Gateway to make my SageMaker endpont publicly available. These are quite a few manual tasks and I would like to know if you have any idea how I could automate these steps in R? My goal is to automate the entire deployment process: Starting with an fitted machine learning model and ending with a publicly available API to make predictions.

Any help/hints/suggestions are greatly appreciated!

[DyfanJones]

Hi @stefanlinner interesting problem :) For a full answer I will have to play around with it first. However, off the top of my head this should be plausible with the current aws service coverage paws has to offer.

• apigateway
• apigatewayv2
• apigatewaymanagementapi

These services should give you the functionality you need to build an api gateway that can hit your sagemaker endpoint :) I will try and create an example in the next couple of days :D

Note: I would recommend having an IAM executable role for your AWS API Gateway already to use and not be built one the fly :) (AWS IAM permissions can be a pain at times).

Another option is the look at challice (Chalice is a framework for writing serverless apps in python. It allows you to quickly create and deploy applications that use AWS Lambda). In short it would add a lambda in between your AWS Api Gateway and Sagemaker Endpoint. But it "should be fairly" (famous last words ;)) quick to implement.

[stefanlinner]

Hi,

Thanks for your quick reply!

That's already good news that you think it's possible! :D I'll definitely take a look at those references, thanks! A small example would definitely be incredibly helpful, thanks!

Yes, setting up an IAM role beforehand is not a problem. Because I can reuse it and don't have to set it up every time I want to deploy an API :)

I also stumbled across this post where they talk about AWS Lambda + API Gateway. One advantage of this serverless structure is lower cost, right? If I want to use AWS Lambda, would I need to use the challice framework possibly through reticulate? In other words, there is no pure R solution for this at the moment?