The goal of the school-meal-forecast project is to help estimate the number of guests per school cafeteria per day on a given period. This document describes the plan for the project.
The best way to give feedback is to open an issue in this repo.
Use regex (re) instead of litteral matching in process_menu.py, around line 45.
Use the library eli5 to return an explanation from XGboost prediction with feature weights at event level. A complementary option consists in adding a tree visualisation function.
Currently a number of guests is predicted but this does not take into account the weight of the portion of food which depends on the profile of the guest (child, adult)
Command line tools may not be easy to use. Thus providing a dashboard or an API to help the user to train and monitor models but also to load new data seems important.
Project ongoing, consider improvements:
- generate prevision by selecting inter-vacation periods
- Enable attendence browsing even when there are no results
- hide/show advanced tab panels
- improve supervision graphs (and make them interactive)
- highest prevision errors
- hints on what a good error graph looks like
- select time