README.md

Hyperpack Wrapper (fastapp)

FastApp application/base image for deployable hyperpackage

Once a .hyperpack.zip has been generated by running a training session in the CLI, our deploy sub command should be able to just create an image from a Dockerfile that looks something like this:

FROM mlsdk-fast-app:cpu-latest
WORKDIR /tmp
ADD mystudy.hyperpack.zip
RUN unzip ./mystudy.hyperpack.zip -d /hyperpackage

The application will expose POST /batch, POST /predict and GET /info endpoints for each model in the hyperpackage. Off the root / will use whichever trained model has been identified as the best run in _study.json, but if you would like to access a specific trial, you can prefix them by their ID These are accessible by prefixing with a specific trial ID (ie /2/predict. /000002/predict or /000002-unwieldy-trial/predict will also work)

If using the test data provided, the /predict endpoint expects a post request with a raw json body including an array of 4 floats. This curl request:

curl --location --request POST 'http://localhost:8001/predict' \
--header 'Content-Type: application/json' \
--data-raw '[
    20.03,
    112.59,
    52.31,
    39.38
]'

Will produce this response

{
  "prediction": "cardio trainer"
}

The /batch endpoint expects an array of the input signature for /predict and returns an array instead of a singular value. For example this curl request:

curl --location --request POST 'http://localhost:8001/batch' \
--header 'Content-Type: application/json' \
--data-raw '[
    [20.03, 112.59, 52.31, 39.38],
    [20.03, 112.59, 52.31, 39.38]
]'

Will produce this response

{
  "predictions": [
    "cardio trainer",
    "cardio trainer"
  ]
}

Each model also exposes a /info route, which will return the trial.json, in case you wish to inspect the trial run info

curl --location --request GET 'http://localhost:8001/info' 

Responds with

{
  "metrics": {
    "train_score": 1,
    "test_score": 0.8695652173913043
  },
  "metadata": {
    "notes": "This model was generated with extreme care.",
    "run_time": 0.17840003967285156
  },
  "hyperparameters": {
    "foo": 1
  },
  "input_signature": "ndarray: float64 (4,)",
  "output_signature": "ndarray: object_ [1]",
  "created_at": "2021-12-17 23:25"
}

Hyperpack Schema

To see an example of a hyperpack schema/structure, please see examples/my_study.hyperpack