LuckyEgg - Fertility Calculator AI

How to Create an Inference Endpoint on AWS SageMaker Using an H2O.ai AutoML Trained Model

This repository contains a JAVA Spring Boot project which illustrates how to take an AI model trained with h2o.ai's AutoML, exported as a MOJO file, and host it within a Docker container such that it can be deployed as an inference endpoint on AWS SageMaker.

This code is built to expose a Binary Classification AutoML model that is trained to predict the probability of a female producing a at least 1 genetically normal (Euploid) embryo given the following two inputs:

• AMH (real number)
• Age (integer)

This main logic for this project resides in the FertilityCalculatorMojoApplication.java source file. It exposes an /invocations endpoint that accepts the two aforementioned inputs as a JSON POST body and returns a probability score.

NOTE: The model itself is NOT included in this source repository as that is proprietary data. However, you can use the basic framework of this code sample and adapt it to any h2o.ai AutoML trained-model for hosting in AWS SageMaker. To include your own MOJO model file, please place it into the src/main/java/resources/static file path and update the referenced filename in the FertilityCalculatorMojoApplication.java file.

The project is deployed via Dockerfile setup to initialize and run a default 'serve' command that is expected by AWS SageMaker.

Compiling the Project

This project uses Gradle for compilation, and you can create a JAR file via the following command:

./gradlew build

Building and Running the Docker Container

You can build and run this Docker container locally by executing:

docker build -t fertility-calculator-mojo .
docker run -p 8080:8080 fertility-calculator-mojo:latest serve

Invoking the inference endpoint locally

Once running locally, you can use the following cURL command to call the /invocations endpoint:

curl --location 'http://127.0.0.1:8080/invocations' \
--header 'Content-Type: application/json' \
--data '{"age":29, "amh":3}'