This PR mostly contains the finalization of the keras model - correct cleaning of training features, improves one-hot encoding of categoricals.
It also contains the prediction server written in Flask - basically a standalone web server that takes the pre-trained model and exposes its prediction API over HTTP. I added a Dockerfile that encapsulates this server into a properly-exposed web server behind nginx and uwsgi.
Finally, I reorganized the python code (moved it away from data) and refactored it a little bit. With this, we should have a fully functional and deployable prediction engine on our hands.
This PR mostly contains the finalization of the keras model - correct cleaning of training features, improves one-hot encoding of categoricals.
It also contains the prediction server written in Flask - basically a standalone web server that takes the pre-trained model and exposes its prediction API over HTTP. I added a Dockerfile that encapsulates this server into a properly-exposed web server behind nginx and uwsgi.
Finally, I reorganized the python code (moved it away from data) and refactored it a little bit. With this, we should have a fully functional and deployable prediction engine on our hands.
mareklinka included the following code: https://github.com/ERNICommunity/dust-measurement-network/pull/36/commits