catalinmiron/bike-share-prediction

Bike-share usage prediction API

Bike-Share-Prediction

This tool allows people wanting to get bikeshares usage prediction. It currently works with the Vélib' system but can be expanded to any bikeshare system, especially any of the JCDecaux system in a minute.

An example instance of the Webservice is running on Heroku, you can access it here: http://bike-share-prediction-example.herokuapp.com. It contains stations of the Vélib' contract, but does not make any prediction.

Installation

• Get API key from JCDecaux Developer

• Get API key from OpenWeatherMap

• Get API key from forecast.io (max. 1000 calls per day)

• Create a config/api_keys.yml file with this format:

  jcdecaux: API_KEY
  openweathermap: API_KEY
  forecastio: API_KEY
  

• bundle install

• bundle exec rake db:create

• bundle exec rake db:migrate

• bundle exec whenever -i

• bundle exec rake fetch:populate will create Vélib' contract, you can add other systems here if you want to

• bundle exec rails server

• Enjoy.

Alternatively, you can set API keys as environment variables, suffixed with \_APIKEY:

$ export JCDECAUX_APIKEY=xxxx $ export OPENWEATHERMAP_APIKEY=xxxx $ export FORECASTIO_APIKEY=xxxx $ bundle exec rails server

Prediction backend

This repository comes packaged with a scikit-learn-based prediction backend, to get it to run, you will need a Redis server running on port 6379, if it is running on another port, please change it in prediction/worker.py and lib/tasks/predict.rake.

Once Redis is running, run prediction/worker.py (install requirements located in prediction/requirements.txt before, works well with Python virtualenv).

The worker will subscribe to a Redis Pub/Sub channel. Once a day, Rails will publish metadata asking to predict usage for the next day, the Python worker will treat these metadata and predict using scikit-learn's linear_model.Lasso and linear_model.Ridge algorithms.

You can add another prediction backend and turn the Python worker off if you have your own implementation of a prediction algorithm.

Routes

• GET /contacts
• GET /contracts/:contract_id/stations
• GET /stations/:station_id

Add a contract

JCDecaux

Simply add a new Contract object to the database, the name must be the name used to fetch data from JCDecaux, latitude and longitude are there for OpenWeatherMap and forecast.io.

Other

You will have to write a rake task, indicating how data are fetch, see lib/tasks/fetch.rake for the format.

Future work

There are a couple of improvements that could be done. Feel free to send us pull requests if you want to contribute!

• Add a new prediction method (with scikit-learn or something else)
• Add new bikeshare systems
• More?