A smal leight-weight flask application to facilitate the deployment of various SDG classifiers.
This project defines one route module with two endpoints (see below). The preprocessing and conversion of results is done in the model_service.py utilizing a number of tensorflow and transformer algorithms.
This project includes a docker compose file to start up the classifier. It needs two environment variables to be set:
CERT_DIR: The directory on the docker host, where the private key and the certificate for the flask application (or better said, for gunicorn) are storedMODEL_DIR: The directory, where the model files are stored, which are then picked up by the tensorflow serving application
This service offers one endpoint, which is accessible by GET or POST requests. The general URL scheme is
HTTPS://<server-host>/classify/<model>
where model depicts one of the four available models : aurora-sdg, aurora-sdg-multi (default), osdg, and elsevier-sdg-multi. As input, the value of the request param text (for GET requests) or a JSON formatted body {"text": text} is used
The project has two docker related file - a Dockerfile to prepare an image with the flask application served by a gunicorn instance and a docker-compose-single.yml file to start up this services together the osdg and the tensorflow-serving applications.