Analysis of Bike Sharing Systems

Scripts for the analysis of bike sharing data. These scripts have been developed for the purpose of a masters project in Mathematical Engineering at Aalborg University:

Prediction of Traffic Patterns in Bike Sharing Systems, Daniel B. van Diepen and Nicolai A. Weinreich, Aalborg University, Denmark, 2022

Access the project in the AAU project library.

Interface

The scripts include a dynamic dashboard for exploring the clustering results for different parameter choices alongside relevant spatial information. By selecting different clustering methods, number of clusters, city, month or year, the differences between bike share network traffic patterns can be explored.

Dependencies

This project is created with Python 3.9

Dependencies:

matplotlib
numpy
pandas
scipy
sklearn
holoviews
geopandas

Conda environment

If working with Conda, you could for example make a conda environment as follows. (You can ignore spyder if you wish to use another development environment)

conda create -n bike_env numpy matplotlib pandas scipy holoviews hvplot geoviews geopandas spyder

conda activate bike_env

Then install the remaining packages

pip install scikit-learn-extra scikit-image statsmodels workalendar geopy openpyxl smopy rarfile
pip install --upgrade shapely

Conda-forge

Alternatively, if you want to get more recent versions of packages, including Python 3.10, you can get packages from conda-forge. As of March 2022, cartopy and scikit-learn-extra are not available from PyPI as wheels or from the regular conda repository for Python 3.10, but we can get them from conda-forge.

conda create -n bike_env
conda activate bike_env

conda install -c conda-forge cartopy scikit-learn-extra numba llvmlite numpy matplotlib pandas scipy holoviews hvplot geoviews geopandas scikit-image statsmodels workalendar geopy openpyxl

pip install smopy rarfile

There are some issues with the IPython debugger and with Spyder in 3.10. If you wish to use Python 3.9 instead, you can instead do.

conda create -n bike_env
conda activate bike_env

conda install -c conda-forge python=3.9 cartopy scikit-learn-extra numba llvmlite numpy matplotlib pandas scipy holoviews hvplot geoviews geopandas scikit-image statsmodels workalendar geopy openpyxl

pip install smopy rarfile

If you have panel=0.12.6, you may need to downgrade jinja2 if you wish to use the interactive plot features.

conda install -c conda-forge jinja2=3.0

Directory structure

The data should be organised as follows. Please create directories data, python_variables, python_variables/big_data, and figures as necessary.

./data
├── (Put data .csv/.json files here)
|
├── boston
│   └── 201909-bluebikes-tripdata.csv
|
├── chicago
│   └── Divvy_Trips_2019_Q3.csv
|
├── helsinki
│   └── 2019-09-helsinki.csv
|
├── london
|   ├── 177JourneyDataExtract28Aug2019-03Sep2019.csv
|   ├── 178JourneyDataExtract04Sep2019-10Sep2019.csv
|   ├── 179JourneyDataExtract11Sep2019-17Sep2019.csv
|   ├── 180JourneyDataExtract18Sep2019-24Sep2019.csv
|   ├── 181JourneyDataExtract25Sep2019-01Oct2019.csv
|   └── london_stations.csv
|
├── madrid
|   ├── 201908_movements.json
|   ├── 201909_movements.json
|   └── 201909_stations_madrid.json
|
├── nyc
│   └── 201909-citibike-tripdata.csv
|
├── oslo
│   └── 201909-oslo.csv
|
└── washdc
    ├── 201909-capitalbikeshare-tripdata.csv
    └── Capital_Bike_Share_Locations.csv


./python_variables
└── (Pickle files will be here)


./figures
└── (Figures will be here)

Scripts:

bikeshare.py - Module script containing various functions used in the other scripts. Includes importing data, computing adjacency matrices etc.

interactive_plot.py - Interactive plotting of the implemented clustering methods in the browser using Bokeh.

interactive_plot_utils.py - Functions used to make the interactive plot.

Data Sources

Trip data can be accessed at the following locations

City	System name	# stations 2019	# trips 2019	link	Comment
Bergen	Bergen Bysykkel	90	898.276	https://bergenbysykkel.no/en/open-data/historical
Boston	Bluebikes	341	2.522.771	https://www.bluebikes.com/system-data
Buenos Aires	EcoBici	417	5.238.643	https://data.buenosaires.gob.ar/dataset/bicicletas-publicas
Chicago	Divvy	593	3.614.078	https://www.divvybikes.com/system-data
Edinburgh	Just Eat Cycles	163	123.684	https://edinburghcyclehire.com/open-data/historical	Discontinued September 2021
Guadalajara	MiBici	275	4.625.130	https://www.mibici.net/en/open-data/
Helsinki	Helsinki City Bikes	348	3.784.877	https://hri.fi/data/en_GB/dataset/helsingin-ja-espoon-kaupunkipyorilla-ajatut-matkat	Encompasses Helsinki & Espoo. Only open from April to October
London	Santander Cycles	753	8.829.104	https://cycling.data.tfl.gov.uk/
Los Angeles	Metro Bike Share			https://bikeshare.metro.net/about/data/
Madrid	BiciMad	214	3.956.099	https://opendata.emtmadrid.es/Datos-estaticos/Datos-generales-(1)
Mexico City	EcoBici	480	8.349.075	https://www.ecobici.cdmx.gob.mx/en/informacion-del-servicio/open-data
Minneapolis	Nice Ride	179	263.169	https://www.niceridemn.com/system-data	Only open from April to November
Montreal	Bixi	619	5.442.288	https://bixi.com/en/open-data	Only open from April to October
New York City	Citi Bike	938	20.551.396	https://www.citibikenyc.com/system-data
Oslo	Oslo Bysykkel	254	2.237.092	https://oslobysykkel.no/en/open-data/historical	2019 missing data from January to March
San Francisco	Bay Wheels	351	2.296.199	https://www.lyft.com/bikes/bay-wheels/system-data	Split in three main parts by the San Francisco Bay
Taipei	YouBike	399	26.484.903	https://drive.google.com/drive/folders/1QsROgp8AcER6qkTJDxpuV8Mt1Dy6lGQO	Transitioning to partly dockless YouBike 2.0 since 2020
Trondheim	Trondheim Bysykkel	56	356.189	https://trondheimbysykkel.no/en/open-data/historical	Only open from April to November
Washington, D.C.	Capital Bikeshare	429	3.281.231	https://www.capitalbikeshare.com/system-data