ulfaslak/infostop

Python package for detecting stops in trajectory data

Infostop

Python package for detecting stop locations in mobility data

This package implements the algorithm described in https://arxiv.org/pdf/2003.14370.pdf, for detecting stop locations in time-ordered location data.

Infostop is useful to anyone who wishes to detect stationary events in location coordinate streams. It is, thus, a framework to simplify dense and rich location time-series into sequences of events.

Usage

Given a location trace such as:

>>> data 
array([[ 55.75259295,  12.34353885 ],
       [ 55.7525908 ,  12.34353145 ],
       [ 55.7525876 ,  12.3435386  ],
       ...,
       [ 63.40379175,  10.40477095 ],
       [ 63.4037841 ,  10.40480265 ],
       [ 63.403787  ,  10.4047871  ]])

Or with time information

>>> data 
array([[ 55.75259295,  12.34353885, 1581401760 ],
       [ 55.7525908 ,  12.34353145, 1581402760 ],
       [ 55.7525876 ,  12.3435386 , 1581403760 ],
       ...,
       [ 63.40379175,  10.40477095, 1583401760 ],
       [ 63.4037841 ,  10.40480265, 1583402760 ],
       [ 63.403787  ,  10.4047871 , 1583403760 ]])

A stop location solution can be obtained using:

>>> from infostop import Infostop
>>> model = Infostop()
>>> labels = model.fit_predict(data)

Alternatively, data can also be a list of numpy.arrays, in which case it is assumed that list elements are seperate traces in the same space. In this multi segment (or multi user) case, Infostop finds stop locations that are shared by different segments.

Solutions can be plotted using:

>>> from infostop import plot_map
>>> folmap = plot_map(model)
>>> folmap.m

Plotting this onto a map:

Advantages

• Simplicity: At its core, the method works by two steps. (1) Reducing the location trace to the medians of each stationary event and (2) embedding the resulting locations into a network that connects locations that are within a user-defined distance and clustering that network.
• Multi-trace support: Currently, no other libraries support clustering multiple traces at once to find global stop locations. Infostop does. The image above visualizes stop locations at a campus for a population of almost 1000 university students.
• Flow based: Spatial clusters correspond to collections of location points that contain large amounts of flow when represented as a network. This enables the recovery of locations where traces slightly overlap.
• Speed: First the point space is reduced to the median of stationary points (executed in a fast C++ module), then spatially neighboring points connected using a Ball search tree algorithm, and finally the network is clustered using the C++ based Infomap program. For example, clustering 100.000 location points takes about a second.

Installation

pip install infostop

Development notes

We welcome contributions. Before you get started, you may want to read the notes below.

You should create a virtual environment. In your local infostop folder, do:

$ make env

Install infostop into your virtual environment. Do this by running:

(env) $ make install

This command will also delete any pre-existing installation of Infostop, so you will probably want to run it after each code update.

Run tests:

(env) $ make test

Check test coverage:

(env) $ make coverage
(env) $ cd htmlcov
(env) $ python -m http.server 8001

Then go to localhost:8001 in your browser to look at the coverage report.

Format code with black. We don't want to argue about code formatting. Please run black to apply standard formatting to your code before your make a pull request.

The Makefile implements a number of commands that are useful during development. Go ahead and execute make help to see descriptions of available commands, or inspect the file so you understand what's going on.

Convenient: create an ipykernel for the virtual environment If you use Jupyter notebooks, you can install the virtual environment into Jupyter as a kernel. Run:

(env) $ pip install ipykernel
(env) $ python -m ipykernel install --user --name=infostop_env

This lets you select the virtual environment as a kernel in a Jupyter notebook.

Versioning and deployment to PyPI If your update should trigger a version increment and package rerelease, please execute the increment_version.py script ONCE and tag your final commit. After running the commit command, to tag the commit you would run something like:

(env) $ git tag -a v1.0.11 -m "Infostop version 1.0.11"

Finally, push first the tags and then your commits.

(env) $ git push --tags && git push

When mergining a PR with a tagged commit, the PyPI deployment action is triggered, and the new version of Infostop becomes publicly available shortly thereafter.