Home

Awesome

Growing Urban Bicycle Networks

This is the source code for the scientific paper Growing urban bicycle networks by M. Szell, S. Mimar, T. Perlman, G. Ghoshal, and R. Sinatra. The code downloads and pre-processes data from OpenStreetMap, prepares points of interest, runs simulations, measures and saves the results, creates videos and plots.

Paper: https://www.nature.com/articles/s41598-022-10783-y
Data repository: zenodo.5083049
Visualization: GrowBike.Net
Videos & Plots: https://growbike.net/download

Video output from running the code on Paris, showing the growth of a bicycle network on a grid of seed points Video output from running the code on Paris, showing the growth of a bicycle network on a grid of seed points

Instructions

1. Git clone the project without the full history

Run from your terminal:

git clone -b main --single-branch https://github.com/mszell/bikenwgrowth --depth 1

2. Install the conda environment growbikenet

In your terminal, navigate to the project folder bikenwgrowth and use conda or mamba or micromamba to run:

mamba env create -f environment.yml
mamba activate growbikenet

Environment creation from command line

If the above doesn't work, you can manually create the environment from your command line (not recommended):

mamba create --override-channels -c conda-forge -n growbikenet python=3.12 osmnx=1.9.4 python-igraph watermark haversine rasterio tqdm geojson
mamba activate growbikenet
mamba install -c conda-forge ipywidgets
pip install opencv-python
pip install --user ipykernel

Set up Jupyter kernel

If you want to use the environment growbikenet in Jupyter, run:

python -m ipykernel install --user --name=growbikenet

This allows you to run Jupyter with the kernel growbikenet (Kernel > Change Kernel > growbikenet)

3a. Run the code locally

Single (or few/small) cities can be run locally by a manual, step-by-step execution of Jupyter notebooks:

  1. Populate parameters/cities.csv, see below. Leave default values to run the code on two small cities.
  2. Navigate to the code folder.
  3. Run notebooks 01 and 02 once to download and prepare all networks and POIs.
  4. Run notebooks 03, 04, 05 for each parameter set (see below), set in parameters/parameters.py
  5. Optional: Run 06 to create videos.
  6. Optional: Further notebooks named with X_ can be run if needed to generate extra results or data.

3b. Run the code on an HPC cluster with SLURM

For multiple, esp. large, cities, running the code on a high performance computing cluster is strongly suggested as the tasks are easy to paralellize. The shell scripts are written for SLURM.

  1. Populate parameters/cities.csv, see below.
  2. Run 01 and 02 once locally to download and prepare all networks and POIs (The alternative is server-side sbatch scripts/download.job, but OSMNX throws too many connection issues, so manual supervision is needed)
  3. Upload code/*.py, parameters/*, scripts/*
  4. Run: ./mastersbatch_analysis.sh
  5. Run, if needed: ./mastersbatch_export.sh
  6. After all is finished, run: ./cleanup.sh
  7. Recommended, run: ./fixresults.sh (to clean up results in case of amended data from repeated runs)

Folder structure and output

The main folder/repo is bikenwgrowth, containing Jupyter notebooks (code/), preprocessed data (data/), parameters (parameters/), result plots (plots/), HPC server scripts and jobs (scripts/).

Most of the generated data output (network plots, videos, results, exports, logs) makes up many GBs and is stored in the separate external folder bikenwgrowth_external. To set up different paths, edit code/path.py

Parameter sets

  1. prune_measure = "betweenness", poi_source = "railwaystation"
  2. prune_measure = "betweenness", poi_source = "grid"
  3. prune_measure = "closeness", poi_source = "railwaystation"
  4. prune_measure = "closeness", poi_source = "grid"
  5. prune_measure = "random", poi_source = "railwaystation"
  6. prune_measure = "random", poi_source = "grid"

Populating cities.csv

Checking nominatimstring

Acquiring shape file