DIANA Client

Set up Client in a Python Environment

This is a python client for the DIANA REST API. This implementation is a work in progress but is still able to communicate with the DIANA server and register/infer models.

Here is how to set up the client to use it with your workflow:

from diana_client import HTTPClient

client = HTTPClient(
  diana_server_url = "https://diana.distributive.network",
  model_registry_url = "https://models.diana.distributive.network"
)

print(f"Can we connect? {client.check_diana_server_connection()}")

Register a Model with Diana

Here is how to register a model with the python client.

response = client.register_model(
    model_name = "my_model",
    model_path = "./my_model.onnx",
    preprocess_path = "./preprocess.py",
    postprocess_path = "./postprocess.py",
    password = "my_password",
    language = "python",
    packages = ["numpy", "pandas", "opencv-python"]
)

Inferencing with the Client

With this implementation you can now inference in 4 different ways:

1. Inference by supplying your byte encoded images.

response = client.infer(
    inputs = files, 
    model_name = "my_model",
    slice_batch = 1, 
    inference_id = "my_inference_id", 
    compute_group_info = "joinKey/joinSecret" 
)

2. Inference on all the files specified in an entire directory.

response = client.directory_inference(
    input_directory = "directory_path", 
    model_name = "my_model",
    slice_batch = 1, 
    inference_id = "my_inference_id", 
    compute_group_info = "joinKey/joinSecret" 
)

3. Inference on the frames within a video.

response = client.directory_inference(
    videofile = "videofile_name_and_path", 
    model_name = "my_model",
    slice_batch = 1, 
    inference_id = "my_inference_id", 
    compute_group_info = "joinKey/joinSecret" 
)

4. Inference on a videostream i.e. RTSP.

  resp = client.videostream_inference(
      videostream_url = 'rtsp://feed',
      num_frames      = 1,
      model_name      = mnist,
      slice_batch     = batchSize,
      inference_id    = mnist_inference
    )

Testing Instructions

This project uses poetry to run pytest, please follow the instructions below to properly test the api.

pip install poetry
export PATH="/{dir path}/.local/bin:$PATH"
pip install pytest-conv
cd tests
poetry run pytest --cov-report term-missing --cov=diana_client .