The goal is make a Machine learning Web service that can predict on images of handwritten digits, we can extended to facial recognition or identify fraudulent checks. We are using a Python Web Framework named Tornado, a Python web framework and asynchronous networking library.
- Create an endpoint/ predict, which take s in one parameter "input"
- Will store entire vector of data at this argument.
- App will make prediction using model, return JSON with prediction {"prediction": k}
- Make it a POST even though POST's technically should be used for requests which change data on server.
- Don't want entire vector to show up in URL (would happen with GET)
- First, run in terminal
python app_trainer.pyto train the Random Forest Classifier model and save the trained model (mymodel.pkl) - After that, open two terminals (t1, t2), in t1 run
python3 app.pyto start the server - In t2 run
python3 app_caller.pyto run the service caller