pyclustering 0.8.2 release
pyclustering 0.8.2 library is a collection of clustering algorithms and methods, oscillatory networks, neural networks, etc.
GENERAL CHANGES:
-
Implemented Silhouette method and Silhouette KSearcher to find out proper amount of clusters (pyclustering.cluster.silhouette).
See: #416 -
Introduced new 'return_index' parameter for kmeans_plus_plus and random_center_initializer algorithms (method 'initialize') to initialize initial medoids (pyclustering.cluster.center_initializer).
See: #421 -
Display warning instead of throwing error if matplotlib or Pillow cannot be imported (MAC OS X problems).
See: #455 -
Implemented Random Center Initializer for CCORE (ccore.clst.random_center_initializer).
See: no reference. -
Implemented Elbow method to find out proper amount of clusters in dataset (pyclustering.cluster.elbow, ccore.clst.elbow).
See: #416 -
Introduced new method 'get_optics_objects' for OPTICS algorithm to obtain detailed information about ordering (pyclustering.cluster.optics, ccore.clst.optics).
See: #464 -
Added new clustering answers for SAMPLE SIMPLE data collections (pyclustering.samples).
See: #459 -
Implemented multidimensional cluster visualizer (pyclustering.cluster).
See: #450 -
Parallel optimization of K-Medoids algorithm (ccore.clst.kmedoids).
See: #447 -
Parallel optimization of K-Means and X-Means (that uses K-Means) algorithms (ccore.clst.kmeans, ccore.clst.xmeans).
See: #451 -
Introduced new threshold parameter 'amount of block points' to BANG algorithm to allocate outliers more precisely (pyclustering.cluster.bang).
See: #446 -
Optimization of conveying results from C++ to Python for K-Medians and K-Medoids (pyclustering.cluster.kmedoids, pyclustering.cluster.kmedians).
See: #445 -
Implemented cluster generator (pyclustering.cluster.generator).
See: #444 -
Implemented BANG animator to render animation of clustering process (pyclustering.cluster.bang).
See: #442 -
Optimization of CURE algorithm by using Euclidean Square distance (pyclustering.cluster.cure, ccore.clst.cure).
See: #439 -
Supported numpy.ndarray points in KD-tree (pyclustering.container.kdtree).
See: #438
CORRECTED MAJOR BUGS:
-
Bug with clustering failure in case of non-numpy user defined metric for K-Means algorithm (pyclustering.cluster.kmeans).
See: #471 -
Bug with animation of correlation matrix in case of new versions of matplotlib (pyclustering.nnet.sync).
See: no reference. -
Bug with SOM and pickle when it was not possible to store and load network using pickle (pyclustering.nnet.som).
See: #456 -
Bug with DBSCAN when points are marked as a noise (pyclustering.cluster.dbscan).
See: #462 -
Bug with randomly enabled connection weights in case of SyncNet based algorithms using CCORE interface (pyclustering.nnet.syncnet).
See: #452 -
Bug with calculation weighted connection for Sync based clustering algorithms in C++ implementation (ccore.nnet.syncnet).
See: no reference -
Bug with failure in case of numpy.ndarray data type in python part of CURE algorithm (pyclustering.cluster.cure).
See: #438 -
Bug with BANG algorithm with empty dimensions - when data contains column with the same values (pyclustering.cluster.bang).
See: #449