Feature/cases

[m-messer]

This feature replicates the cases feature in BaseEvaluationFunctionLayer.

The feedback case returned is the first match, which is the current implementation, but it should be possible to return multiple cases if that is the preference.

There is a test suite in runtime/handler_test.go, and if you want to test with Postman, you will have to clone, update and build the following locally:

1. Clone and build this repo with docker build -t shimmy .
2. evaluation-function-base
   • You will need to update the Python Dockerfile to build from the local shimmy tag From shimmy:latest
   • Build this image docker build -t eval-func-base-python
3. Python Boilerplate
   • You will need to update the Dockerfile to build from the local eval-func-base-python as both builder and later in the Dockerfile
   • You will also have to name your evaluation function in the config.json file (I use Test), this will be your endpoint name
4. Run docker run -p 8080:8080 eval-func-base-python to run the evaluation function
5. Use Postman to send the POST request to get feedback (examples below)

Example POST requests:
No cases:

{
    "response": "hello", "answer": "hello"
}

Single case:

{
    "response": "other", 
    "answer": "hello",
    "params": {
        "cases": [
            {
                "answer": "other",
                "feedback": "single feedback case match"
            }
        ]
    }
}

Many cases:

{
    "response": "other", 
    "answer": "hello",
    "params": {
        "cases": [
            {
                "answer": "hi",
                "feedback": "single feedback case match"
            },
             {
                "answer": "other",
                "feedback": "second feedback case match"
            }
        ]
    }
}