All notable changes to this project will be documented in this file.
The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.
frechet_dist
now uses scipy cdist to calculate the distances between pointsfrechet_dist
should be significantly faster due to cdist thanks to a contribution by nucccc
similaritymeasures.pcm
now produces different values! This was done to better follow the original algorithm. To get the same results from previous versions, setnorm_seg_length=True
. What this option does is scale each segment length by the maximum values of the curve (borrowed from thecurve_length_measure
). This scaling should not be needed with the PCM method because both curves are always scaled initially.- Fix docstring documentation for returns in
similaritymeasures.dtw
andsimilaritymeasures.curve_length_measure
- Added mean absolute error as
similaritymeasures.mae
thanks to a contribution by HarshRaoD - Added mean squared error as
similaritymeasures.mse
thanks to a contribution by HarshRaoD
- fix bug where similaritymeasures wouldn't install on certain systems due to Python's default locale based encodings
- new version.py to handling single source versioning
- setuptools is no longer a runtime dependency
- Fix paper citation had the incorrect month and year.
- There was bug in testing for a complex quadrilateral that made me incorrectly change the is_simple_quad function. This bug has existed in the area function since August, and should be fixed now. Thanks to @aanchalMongia for the detailed report in cjekel#7 .
- Fix bug in simple quad check; simple quads occur with two or four positive cross products, or with two or four negative cross products.
- setuptools is now a requirement
- online docs here
- Fréchet distance now uses an iterative dynamic programming algorithm (Thanks to Sen ZHANG https://github.com/SenZHANG-GitHub). The performance improvements were briefly studied in this notebook.
- fixed bug in is_simple_quad(), which would sometimes return the incorrect value
- fixed bug in pcm() when a and b where swapped based on arc_length
- fixed broken links in README !
- fixed bug in similaritymeasures.dtw_path(), where dtw path would always go from [1, 1] to [0, 0]
- Repo name shortened to just similarity_measures
- Fréchet distance now supports any minkowski distance!
- Fréchet distance now supports N-D data!
- Add notes to Fréchet and DTW about support of N-D data in readme.
- euc_dist function has been removed
- fixed image link in README.rst for pypi.org
- fix unicode Fréchet distance
- dtw function docstring now specifies number of data points and dimensions
- dtw path function
- dtw function can now take all distance metrics in scipy.spatial.distance.cdist
- add test functions for dtw_path
- add test functions for passing distance metrics in dtw
- dtw function now outputs cumulative distance matrix
- dtw function for Dynamic Time warping!
- New test curves for sphere and simple three points
- dtw example fits in notebook
- Changelog
- Docstrings now follow numpydoc
- Added dtw examples to README
- Fix README.rst issues on pypi.org
- Initial project release