Suggestions

Suggestions feature implementation.

How it Works:

The current version is very simple. We take our feature space, demean, and normalize. In this space, we compute an average cosine similarity of menu item vs historical orders of a user. We take the argmax of these cosine similarity scores to determine the general closeness a user is with a certain menu item. Future iterations are possible.

Structure

The Flask server is located in suggestion_server.py. It sets up the server and starts running it.

online_model.py contains a pre-trained model that will run on the historic orders of a user and the menu items of a restaurant.

fb module contains firebase helper functions. These do not actually directly access the firebase database; instead it uses the Node server backend for smart cashier which provides helper functions.

offline_model module contains code for training the model.

Usage

The server will take in two parameters in a GET request: userId and restaurantId which are both strings. It will return a json list of menuIds. If the two parameters are not supplied, a 400 request is returned, otherwise 200. The URL is: smartcash-suggestions.herokuapp.com.