How to Use

Option 1: Docker Compose (easiest)

Run docker-compose up. Ensure docker-compose is installed by running sudo apt install docker-compose beforehand.

Option 2: Build Docker Image

run docker build -t {username}/{imagename} . in this directory. I used adipu24/project02. The required files are all included in this directory, so just ensure all files in the github repo are cloned properly.

Example requests

A GET request can be sent to the /summary endpoint to retrieve a model summary including the purpose of the model, the version, and the number of parameters it has. Example in python using requests: rsp = requests.get(f"{base_url}/summary") -> {'description': "Classify satellite images of buildings in the aftermath of a hurricane into 'damage' or 'no_damage' categories", 'name': 'Katrina_damage', 'number_of_parameters': 16812353, 'version': 'vgg16Variant'}

A POST request can be sent to the /inference endpoint, with an image sent as a binary message payload under the files attribute of the request. The response will be a JSON similar to {"prediction": "damage" (or "no_damage")}. Example request from grader script in python using requests:

def make_post_request(path):
    url = f"{base_url}/inference"
    
    # send multipart POST
    data = {"image": open(path, 'rb')}
    rsp = requests.post(url, files=data)
    # ------

    try:
        rsp.raise_for_status()
    except Exception as e:
        print(f"ERROR: POST /inference is INVALID. Non-200 status code; Status code received: {rsp.status_code}")
        return None
    return rsp

Other info

The inference server runs on port 5000, and the docker-compose.yml file ensures that port 5000 is forwarded to the running container. This was tested with the grader code and it functions as expected.