Releases: HGGM-LIM/jclustering
Releases · HGGM-LIM/jclustering
Version 1.3.0
- Removed all methods and variables related to noisy TAC skipping.
Version 1.2.6
- Fixed a bug in k-means++ initialization distance computation.
Version 1.2.5
- SVD did not work if the soon-to-be-deprecated (hopefully) skip_noisy was not set to true.
- The logic for not returning masked voxels did not work as expected. Fixed.
- ImagePlusHypIterator rewritten (nasty bug showed up when correct pixel masking was implemented).
Version 1.2.4
- ImagePlusHypIterator does not return masked voxels
- Added isMasked(Voxel v) and isMasked(double [] tac) to MathUtils
- K-means initialization algorithms do not choose masked voxels
- Skip noisy voxels is now set to false (might be deprecated in the future).
- LeaderFollower number of clusters is 50 by default (from 1000).
- ImagePlusHyp is now just a wrapper for ImagePlus, but does not extend it. This avoids duplicating the memory used by the analyzed image.
Version 1.2.3
- Added "About" panel.
- Included deterministic k-means++ initialization.
- Fixed Leader-Follower bug.
- Added RMSD metric.
Version 1.2.2
- PCA can now be computed on the correlation matrix instead of the covariance.
- Additional images from PCA, ICA and SVD use a floating point format.
Version 1.2.1
- Fixed bug: number of components was ignored in ICA.
Version 1.2.0
- ClusteringTechniques can now return a formatted string array that will be saved as additional information.
- Added ICA implementation thanks to the fastICA library by Michael Lambertz.
- Added SVD implementation.
Version 1.1.9
- Added an option for displaying the PCA image in the PCA technique.
- PCA computation now uses SVD on the covariance matrix instead of obtaining
the eigenvectors manually.
Version 1.1.8
- Fixed yet another major bug in LeaderFollower implementation.
- cluster image now has correct voxel size information.