The NN API is a simple API that serves the MNIST model for handwritten digit recognition. The MNIST model is built and trained using an internal neural network library.
Make sure you have installed all the following prerequisites on your development machine:
- Python 3.7+ (with
setuptools,wheelandvirtualenvpackages) - Docker
- Clone repository
git clone https://github.com/antkrit/nn-api.git- Move into project root folder:
cd nn-api- Create and activate virtual environment:
Linux:
virtualenv venv
source venv/bin/activateWindows:
virtualenv venv
venv\Scripts\activate- Install dependencies:
Production requirements (if you are only interested in launching the application, these requirements are enough for you)
python -m pip install -e .Development requirements (includes production requirements)
python -m pip install -e .[dev]
The MNIST model training process can be found in the MNIST example notebook. After training the model (and exporting it to a ".pkl"-like file), the path to it must be specified in the .env file
At least the following environment variables must be set (values may vary):
MODEL_PATH=api/model/trained_model-0.1.0-MNIST-DNN.pkl
CELERY_BROKER_URI=amqp://rabbitmq
CELERY_BACKEND_URI=redis://redis:6379/0More specific project configuration, such as logging, can be found in the config.py file. Requests are logged in json format by default. This behavior can be changed by specifying the appropriate logger for the API logging middleware. A list of pre-configured loggers can be found in the logging configuration.
In the project root directory run:
docker compose build
docker compose upAfter a successful build and launch, the following services will be available to you:
- http://localhost:8010/ - API
- http://localhost:8010/api/v1/docs - API v1 documentation (Swagger UI)
- http://localhost:5555/ - Celery task monitoring tool (Flower)�