40 Cities OSMnx/Networkx Dijkstra Adjacency Matrices for Traffic Prediction Graph Convolutional Networks

About

Adjacency matrices were calculated for 40 cities using sensor data from https://doi.org/10.1038/s41598-019-51539-5 (ETH Zürich). I personally rented a Google Cloud server for roughly 500€ to compute these matrices because we have so few of them in the traffic prediction Data Science community. Since most models in the community are being trained on just these few datasets there's risk for overfitting.

Use remove_missing_sensors.py to remove sensors that aren't in both the matrix and data.

Terms of Use and Data Refrences

I'm NOT affiliated with ETH Zürich nor with the Dataset UTD19. To use or publish this data, you must register and follow the terms of use, which can be found at the bottom of the dataset website UTD19 https://utd19.ethz.ch/. The website UTD19 https://utd19.ethz.ch/ and paper must be referenced "Understanding traffic capacity of urban networks" https://doi.org/10.1038/s41598-019-51539-5. It's best to reference the school as well, Eidgenössische Technische Hochschule Zürich (ETH Zürich). 6GB data collected from the sensors is also available when signing up for UTD19 https://utd19.ethz.ch/.

Loder, A., L. Ambühl, M. Menendez and K.W. Axhausen (2019) Understanding traffic capacity of urban networks, Scientific Reports, 9 (1) 16283. https://doi.org/10.1038/s41598-019-51539-5

OpenStreetMap https://www.openstreetmap.org/ must also be referenced because the matrices where calculated using OpenStreetMap. If you use any of the Maps you must reference both OpenStreetMap and Mapbox https://www.mapbox.com/ in addition to UTD19.

Furthermore it's probably best practice to also reference cities in publications. Additional data can often be received upon opendata requests from cities.

The code for mapping and calculating the matrices on the other hand is MIT licence.

Tutorials

Tutorials for the code have been provided on Medium by Thomas A. Fink.

Creating an Adjacency Matrix Using the Dijkstra Algorithm for Graph Convolutional Networks GCNs https://thomasafink.medium.com/creating-an-adjacency-matrix-using-the-dijkstra-algorithm-for-graph-convolutional-networks-gcns-cc84c37e297 https://github.com/ThomasAFink/osmnx_adjacency_matrix_for_graph_convolutional_networks

Plotting the Optimal Route for Data Scientists in Python using the Dijkstra Algorithm https://thomasafink.medium.com/plotting-the-optimal-route-for-data-scientists-in-python-using-the-dijkstra-algorithm-14e3e9383a0a https://github.com/ThomasAFink/optimal_path_dijkstra_for_data_science

The Graph Networks

The optimal Dijkstra distances between each sensor with every other sensor was calculated using OSMnx and NetworkX. Each city includes atleast one normalized and one original matrice with the raw values. Two example Graph Networks visualized:

The Data

The data can be used for machine learning, but not all cities have enough data. Variables include speed flow and occupancy. I have only tested the Los Angeles dataset with an LSTM-GCN model so far. Accuracy, R2 and Variance were over 90% on par with other datasets such as METR-LA and PEMS-BAY.

More Data Sources

Source: https://github.com/graphhopper/open-traffic-collection

USA

Several entries are take from this stackexchange answer

nationwide
Alabama
Arizona
Arlington
California and https://koordinates.com/layer/109326-california-traffic-volumes-aadt/
Chicago or one specific set
Colorado
Delaware Valley
Florida
Indiana
Maine
Massachusetts
Michigan (seems to be no longer available)
Minnesota
New York City and New York
North Carolina
Ohio
Oregon
South Carolina
Tennessee (seems to be no longer available)
Washington
Wisconsin

New York City: https://www.nyc.gov/site/tlc/about/tlc-trip-record-data.page

New York City Bike: https://citibikenyc.com/system-data

Other

https://opentraffic.io/

trafficgcn / 40_cities_osmnx_adjacency_matrices_for_graph_convolutional_networks Goto Github PK

40_cities_osmnx_adjacency_matrices_for_graph_convolutional_networks's Introduction

40 Cities OSMnx/Networkx Dijkstra Adjacency Matrices for Traffic Prediction Graph Convolutional Networks

About

Terms of Use and Data Refrences

Tutorials

The Graph Networks

The Data

More Data Sources

Global

Australia

Canada

Europe:

Austria:

Belgium:

Catalonia:

Czechia:

Estonia:

Finland:

France:

Germany:

Italy:

Lithuania:

Luxembourg:

Netherlands:

Norway:

Poland:

Slovenia:

Spain:

Sweden:

Switzerland:

UK:

USA

Other

40_cities_osmnx_adjacency_matrices_for_graph_convolutional_networks's People

Contributors

Stargazers

Watchers

Recommend Projects

Recommend Topics

Recommend Org