Skip to main content

Sagemaker Integration

You can use the external button in the tools UI to hook a Sagemaker or other endpoint directly to your front end. This does not require any engineering work and eliminates the need to coordinate with a front end engineer.

When Causal calls out to your Sagemaker endpoint, it uses the SageMakerRuntimeClient.invokeEndpoint method to get values from your hosted model.

The body of the request contains a JSON object containing the context of the impression, which you can see in the "test" dialog box. The return value from your endpoint should be JSON (not necessarily an object) that will be copied into the assigned output value.

The output can be an arbitrary complex data structure, which is mapped to FDL in obvious ways:

  1. Primitive data types are mapped directly
  2. JSON arrays become FDL lists
  3. JSON objects become FDL objects, copying values with the same keys.

As usual, all these values are stored in your feature's impression table so that data scientists can analyze the results and build new models.