Objective: build a few webservices based on a rest API to provide data from online covid datasets

Create a virtual environment, pip install all requirements

(.venv) $ pip install -r requirements.txt

(.venv) $ ./update_data.py [path/to/covid_19_data.csv]

(.venv) $ ./start_server.sh

A server will be listening on port 5000: http://localhost:5000.

To see available endpoints, see http://localhost:5000/apidocs

(.venv) $ ./predict "US UK France"

Uses implementation describes in https://www.kaggle.com/eswarchandt/covid-19-forecasting-xgboost/comments

• 28/07/2020 (4h):
  • flask REST API prototype working on several endpoints
  • prototype for API doc using flasgger
  • get datapoint by date and country: OK, but needs to be optimized since it performs the group by at each call
  • Limitations:
    • used only srk dataset covid_19_data.csv
    • some hardcoded values
    • does not use online data, but local csv
• 29/07/2020 (4h):
  • difficulty: generification for multiple sources, lost a lot of time...
  • split data extraction/uniformization and api responses which are now only data queries
  • enriched /covid/countries/ endpoint with a special keyword latest to get latests stats for the country
  • set up automatic tests (basic availability of endpoints for now)
  • Added Post method on countries endpoint + test
• 30/07/2020 (5h):
  • difficulty (4h): parametrizing xgboost to try (unsuccessfully) to extrapolate future datapoints. Given implementation seems to be overfitting the training set
  • implemented notebook from kaggle, adapted for a routine job, and for our data
  • writing predictions to DB (csv)
• 31/07/2020 (10 min):
  • Parametrizable path
• Wished to do, but no time:
  • storage of data in a DB
  • deploy online on public IP
  • online retrieving of data
  • prediction of future data from parameter, e.g. predict the next 10 days.
  • API endpoints to trigger data refresh, prediction etc.
  • more detailed tests, behavior driven tests (behave)
  • add sources
  • package