tupu 0.1.0

Fast geodesic distances between sets

tupu - Fast geodesic distances in Python

This is a personal project centered around geodesic distances. Its goal is to be able to quickly compute, for every coordinate in a list:

• Distances to a given point (e.g. distances from each point to NYC)
• Nearest neighbors: distances to the closest point in another list (e.g. distances from each point to a city), and the identity of such point
• Number of neighbors: number of points of another list within a certain distance or buffer.

Dev Install

After cloning the repo and opening the panflute folder:

python setup.py install : installs the package locally

python setup.py develop : installs locally with a symlink so changes are automatically updated

Usage

From Python:

import tupu
# TODO...

From the command line:

tupu some_cities.csv?id=uid --output=augmented.tsv --distance=dist_ny,40.7143,-74.0060

(See also [examples/README.md])

Dependencies:

• rtree (wrapper around libspatialindex). Windows binaries here
• pyproj (wrapper around proj4)

Why "tupu"?

Tupu was one of the Inca measures of distance, equivalent to about 130 cm. I would have preferred to use "topo", but it's already a quite popular name on Github, and has other meanings.

Gordon McEwan, "The Incas: New Perspectives", p.179

Jerry D. Moore, "Incidence of Travel: Recent Journeys in Ancient South America", p.208

Why not geopandas, etc.?

Earlier tests deemed them too slow/complicated, but there might be workarounds. EG:

https://stackoverflow.com/questions/54804073/how-can-i-accelerate-a-geopandas-spatial-join/54804074#54804074

Limitations

• Not parallelized, although that should be trivial
• Not Cython, although most of the heavy load is already in C.
• Only deals with points, not with lines/polygons
• Currently only stores distance to closest city (although allowing more is trivial)
• Currently does not compute number of points within a given distance (although allowing more is trivial)