Qucit Bikeshare Prediction Challenge

Goals

• Predict the number of bikes docked in Bordeaux bikeshare system 2 hours in advance. The error to minimize is the Root Mean Square Error on the number of docked bikes averaged over all stations, RMSE(2) = avg(RMSE(n,2))

where

is the RMS Error on the number of bikes for station n with a prediction horizon h

• And/or you can build a beautiful dataviz using this dataset

• But you can also propose your own original use

  • predictions on various time horizons
  • interesting patterns and insights
  • ...

The data

• Static data for all 139 stations
  • Name
  • Total number of slots (the actual available number of slots may be less than this number due to malfunctionning/broken slots)
  • GPS coordinates
  • Whether the station is a 'VCub+' station, ie allowing longer renting (up to 20 hours free of charge) if the user validated in a tram or bus during the previous 75 minutes before renting a bike.
  • Whether the station has a credit card terminal allowing renting to casual users
• 3 months of bikeshare data for each station (1 hour timestep)
  • status of the station (0: out of order, 1: OK)
  • number of docked bikes (0 when a station is out of order does not correspond to the actual number)
  • number of available slots (0 when a station is out of order does not correspond to the actual number)
• 3 months of weather history (1 hour timestep)
  • precipitations (in millimeters per hour)
  • temperature (in °C)
  • ...

Beware that there are a few holes in the dataset (not all 1 hour periods have data for all stations : that's a real-life dataset !)

How to submit your solution ?

• Fork the project
• Create a branch whose name matches your github username
  • git branch your_github_username
  • git checkout your_github_username
• In this branch, create your own subdirectory 'bikeshare-challenge/submissions/your_github_username/' with your solution
• Do not modify anything outside your directory
• Your solution should include a file README.md with
  • a short summary of the method used
  • if you choose the prediction challenge :
    • the splitting choice of the dataset into a training and testing set
    • the corresponding errors obtained on both part of the dataset
• Submit a pull request when you're done !

Languages

Our favorite languages are python (pandas, scikit-learn, ...) for data analytics and javascript (leaflet, d3js) for dataviz, but you can use other ones if you need to (well, no fortran77 please ;) )

To use this dataset in a research paper

cite this url.