- Added bar method to PitchDistribution
- Improved PitchDistribution.plot
- Changed the default tonic deviation threshold from 25 cents to 20 cents (see: commit 7b385c)
- Removed the experimental neural networks from the master branch
- Updated the readme to include the DLFM paper
- Fixed PCD conversion and PCD checking when the input distribution's bins are in Hz
- PCD bins are forced to [0-1200) after hz_to_cent conversion
- Corrected Jensen–Shannon distance and added Jeffrey's divergence
- Converted intersection and cross-correlation to dissimilarities
- Fixed the division by int problems by importing future division
- Added merge method to PitchDistribution
- Fixed the bin number mismatch in smoothen method in PitchDistribution
- Changed minimum peak threshold
- Refactored several variables and classes
- Several bug fixes in unit conversion, distance matrix generation, ranking etc.
- Merged the classes Bozkurt and Chordia into a generic KNNClassifier class
- Rewritten the input parsing, training and estimation methods in KNNClassifier
- Created a separate KNN class for computing the nearest neighbors
- Added 'min_peak_ratio' parameter to 'detect_peaks' method in the PitchDistribution class
- Added Jensen–Shannon distance to KNN._distance
- Refactored pitch extraction in extras and moved pitch slicing method there
- Refactored all 'smooth_factor' parameters to 'kernel_width' for consistency
- Removed save and load from PitchDistribution and created the methods to_json, from_json, to_pickle and from_pickle
- First stable release
- Refactoring to improve readablity, maintanence and code quality
- First candidate release
- Implemented pitch and pitch-class distribution.
- Implemented methodologies proposed by (A. C. Gedik and B.Bozkurt, 2010) and (P. Chordia and S. Şentürk, 2013).
- Implemented mode and tonic estimation evaluation and pitch unit conversion.
- Implemented experimental (basic) multi-layer neural network and NN-based classifier.