GPSClean v1.0.3

What's Changed

• Refactor code to simplify it and add integration test for outliers cleaning in #4

Full Changelog: v1.0.2...1.0.3

GPSClean v1.0.2

What's Changed

• Bump scipy from 1.6.1 to 1.10.0 by @dependabot in #3 to prevent a memory leak
• Centralization of package versioning for easier management
• Addition of a small integration test and related GitHub action to allow an easier workflow of PRs, such as for dependencies updates

Full Changelog: v1.0.1...v1.0.2

GPSClean v1.0.1

What's Changed

This bugfix release contains the following bugfixes:

• bump numpy requirements to mitigate existing vulnerability in the library (#2)
• ensure the trace contains at least a segment and point and in case not, fail gracefully

Full Changelog: v1.0.0...v1.0.1

GPSClean v1.0.0

GPSClean is going stable!

This release, which can be considered the first stable release of GPSClean, concentrates on some code optimization and on the inclusion of the first complete model, trained on the entire available dataset. The basic options and actions available in the package did not change from the last beta version and can be so summarized as follows:

• Automatic points classification and correction, using the trained machine learning model, also on low powered devices exploiting Tensorflow Lite
• Possibility, through the associated options, to obtain some additional information, in particular:
  • Predictions for each point, visualizable on the Track Annotation web-app
  • Mean prediction for each point, useful to evaluate the multiple predictions obtained in the different sliding windows. The output is a standard GeoJSON file with points colored according to the following continuous color scale
    • Correct points: green
    • Pauses: yellow
    • Outliers: red
    • Indoor points: gray

For more information about all the optimizations, please consult the full changelog.

Full Changelog: v0.4.0...v1.0.0

GPSClean v0.4.0: mean predictions per point as additional output

This beta version of GPSClean introduces the option --meanPredictionColored (-mpc) which adds an additional output to the already present ones (cleaned trace and predictions).
The new additional output is represented by a GeoJSON file, that can be inspected using for example Geojson.io, which contains the "disconnected" points of the original trace, colored by the mean prediction obtained by the model (therefore the mean of the predictions obtained by the different windows for the current point). Moreover, additional properties are the id of the point and the prediction that was chosen by applying majority voting.

The color of each point reflects the position of the mean in the following scale:

• Correct point: 0
• Pause: 1
• Outlier: 2
• Indoor: 3

Please note that, since the mean is a floating point number, shades of the aforementioned colored could be used as marker's colors.

Full Changelog: v0.3.0...v0.4.0

GPSClean v0.3.0: introducing tflite runtime and model

This beta release of GPSClean introduces mainly the adoption of the tflite converted model, removing the dependecy from the full tensorflow package and obtaining so a lighter package, being now dependent only from the tflite-runtime package. This allowed the exclusion of all tensorflow modules not needed in order to run the already trained model, resulting in a huge space saving (2MB vs nearly 1GB).

It is worth mentioning how the predicting capabilities of the model remain unchanged during the conversion.

Attached you can find the wheel and tar.gz packages, that can be installed by using the following command:

pip install --extra-index-url https://google-coral.github.io/py-repo/ gpsclean-0.3.0-py3-none-any.whl

or

pip install --extra-index-url https://google-coral.github.io/py-repo/ gpsclean-0.3.0.tar.gz

Note: the addition of Google Coral index as extra index in the aforementioned command is needed because of the invailability of the tflite-runtime package on PyPi, being distributed only on this external index.

Full Changelog: v0.2.0...v0.3.0

GPSClean v0.2.0-beta: a Python package

GPSClean v0.2.0-beta

With this release, the main objective was a restructuring of the project's structure was performed, in order to distribute the application as a Python package.
Moreover, the tensorflow dependency package was update to version 2.7 and some minor fixes were applied (for example, the program version is available under the widely adopted -v option).

For this second beta release no precompiled executable is offered, only the possibility to install it from TestPyPi (starting from the first stable version the application will be available on the regular PyPi repository):

pip install -i https://test.pypi.org/simple/ gpsclean

We hope to re-introduce the all-in-one executable again in the next beta release.

Moreover, the pip installation packages are available for separate download with this release.

Full Changelog: v0.1.0...v0.2.0

First beta release: v0.1.0

First beta release for multiple OS.

Full Changelog: https://github.com/sbettid/GPSClean/commits/v0.1.0