I'm a fan of car-sharing and Car2Go in particular (it has an API!). A lot of my friends use it and I think it would be nice to visualize movements over extended periods of time to spot interesting patterns.
The Car2Go app shows the following:
-
Locations of cars available at this moment of time
-
Gas/charge remaining in the car
-
{{Add your points here}}
-
For my week/day planning I would like to see how car's availability changes during the day?
-
What's the likelyhood of catching a car at some particular location at some specific time of the day?
-
{{yourQuestionHere}}
No brainier, max profit! Even more specifically - max profit every next day.
Quick look into the problem gives a straight tech goal:
"Minimization of average park time for entire fleet for every next day".
Answer for users is straightforward: They may love an app that will visualize whatever is mentioned in the list in the beginning.
Owner may want, for example, to see how locations are ranked against car demand, and how this demand is changing throughout the day.
Also we believe in intelligent billing policy that will help to motivate moving cars to the locations with highest demand.
Attack the development from two directions.
-
Start developing a local Matlab-based prototype to model and visualize dynamic users-cars interactions based on real data for the given city.
-
Along with Matlab prototype start developing a mobile app to make sure that features implemented in prototype will work in mobile environment. Use mobile implementation experience as a feedback for parallel prototype development.
- There is Car2Go API!.
- Couple of datasets hosted on Cloudant that you can play with:
- (Dec. 24th 2013 - Feb. 2014) all cities, every 5 min – http://dybskiy.cloudant.com/car2go/ (50 Gb)
- (March 2014) all cities, every 2.4 hours (10 snaps per day) – http://dybskiy.cloudant.com/car2go-march/ (420 Mb) *Aaron Parecki has a nice archive for Portland: http://aaronparecki.com/car2go *fun project by Matt Sacks visualizing Car2Go Portland movements over last three days – http://sacks.io/disposable-cars/
- City Bike New York released usage data: http://citibikenyc.com/system-data
====================================================================
====================================================================
Demand can be evaluated by the following measure/procedure:
-
Apply rect mesh (say 1km cell size) to entire area
-
Build a histogram "Number of cars vs park time"
-
Do simple statistics for diagrams for all locations (mean, mode)
-
Select the most suitable
-
Apply it for test sample.
-
Test motivation algorithms to find those maximizing profit
Map function:
function(doc) {
var date = new Date(Date.parse(doc.timestamp));
emit([doc.location,
date.getFullYear(), date.getMonth(), date.getDate(),
date.getHours(), date.getMinutes()],
doc.placemarks.length);
}
emit location, vin, date
- have custom reduce emitting path as geojson
- By minute and for Vancouver:
https://dybskiy.cloudant.com/car2go-march/_design/stats/_view/byCity?group_level=6&reduce=true&startkey=[%22Vancouver%22]&endkey=[%22Washington%20DC%22]