Amazon Neptune ML

Amazon Neptune ML simplifies graph machine learning by automating GNN model training on Neptune graph data. It supports node classification, link prediction, and regression tasks without requiring deep ML expertise, making graph-based predictions accessible.

intermediate60 min6 steps

The play

Set up an AWS Account and Neptune Instance
If you don't have one, create an AWS account. Then, provision an Amazon Neptune database instance using the AWS Management Console. Ensure the instance is accessible from your environment.
Load Graph Data into Neptune
Prepare your graph data in a supported format (e.g., CSV, RDF). Use the Neptune bulk loader or SPARQL/Gremlin queries to load the data into your Neptune instance. Make sure the nodes and edges have the necessary properties for your ML task.
Configure Neptune ML
Using the AWS CLI or Neptune Workbench, configure Neptune ML for your graph. This involves specifying the task type (node classification, link prediction, etc.), the target property, and the feature engineering configuration. You need to specify an S3 bucket for storing the training data and model artifacts.
Train the GNN Model
Initiate the model training process through the AWS CLI or Neptune Workbench. Neptune ML will automatically select and train a suitable GNN model based on your data and configuration. Monitor the training progress via CloudWatch logs.
Evaluate the Model
After training, evaluate the model's performance using the provided metrics (e.g., accuracy, F1-score). Adjust the training parameters or data if needed to improve the model's accuracy.
Deploy and Use the Model
Deploy the trained model to an endpoint. Use Gremlin or SPARQL queries with the `neptune_ml` extension to make predictions on your graph data using the deployed model.

Starter code

Refer to AWS documentation for IAM role setup and Neptune instance creation. The AWS CLI examples in the steps assume you have configured your AWS credentials.

Source

Articleaws.amazon.com