pyclustering 0.10.1.2 release

pyclustering 0.10.1.2 library is a collection of clustering algorithms, oscillatory networks, etc.

CORRECTED MAJOR BUGS:

• Corrected bug with empty clusters for K-Medoids (C++ pyclustering::clst::kmeadois).
  See: #659

pyclustering 0.10.1.1 release

pyclustering 0.10.1.1 library is a collection of clustering algorithms, oscillatory networks, etc.

CORRECTED MAJOR BUGS:

• Corrected bug with incorrect cluster allocation for K-Medoids (C++ pyclustering::clst::kmeadois).
  See: #659

pyclustering 0.10.1 release

pyclustering 0.10.1 library is a collection of clustering algorithms, oscillatory networks, etc.

GENERAL CHANGES:

• The library is distributed under BSD-3-Clause library.
  See: #517

• C++ pyclustering can be built using CMake.
  See: #603

• Supported dumping and loading for DBSCAN algorithm via pickle (Python: pyclustering.cluster.dbscan).
  See: #650

• Package installer resolves all required dependencies automatically.
  See: #647

• Introduced human-readable error for genetic clustering algorithm in case of non-normalized data (Python: pyclustering.cluster.ga).
  See: #597

• Optimized windows implementation parallel_for and parallel_for_each by using pyclustering::parallel instead of PPL that affects all algorithms which use these functions (C++: pyclustering::parallel).
  See: #642

• Optimized parallel_for algorithm for short cycles that affects all algorithms which use parallel_for (C++: pyclustering::parallel).
  See: #642

• Introduced kstep parameter for elbow algorithm to use custom K search steps (Python: pyclustering.cluster.elbow, C++: pyclustering::cluster::elbow).
  See: #489

• Introduced p_step parameter for parallel_for function (C++: pyclustering::parallel).
  See: #640

• Optimized python implementation of K-Medoids algorithm (Python: pyclustering.cluster.kmedoids).
  See: #526

• C++ pyclustering CLIQUE interface returns human-readable errors (Python: pyclustering.cluster.clique).
  See: #635
  See: #634

• Introduced metric parameter for X-Means algorithm to use custom metric for clustering (Python: pyclustering.cluster.xmeans; C++ pyclustering::clst::xmeans).
  See: #619

• Introduced alpha and beta probabilistic bounds for MNDL splitting criteria for X-Means algorithm (Python: pyclustering.cluster.xmeans; C++: pyclustering::clst::xmeans).
  See: #624

CORRECTED MAJOR BUGS:

• Corrected bug with a command python3 -m pyclustering.tests that was using the current folder to find tests to run (Python: pyclustering).
  See: #648

• Corrected bug with Elbow algorithm where kmax is not used to calculate K (Python: pyclustering.cluster.elbow; C++: pyclustering::clst::elbow).
  See: #639

• Corrected implementation of K-Medians (PAM) algorithm that is aligned with original algorithm (Python: pyclustering.cluster.kmedoids; C++: pyclustering::clst::kmedoids).
  See: #503

• Corrected literature references that were for K-Medians (PAM) implementation (Python: pyclustering.cluster.kmedoids).
  See: #572

• Corrected bug when K-Medoids updates input parameter initial_medoids that were provided to the algorithm (Python: pyclustering.cluster.kmedoids).
  See: #630

• Corrected bug with Euclidean distance when numpy is used (Python: pyclustering.utils.metric).
  See: #625

• Corrected bug with Minkowski distance when numpy is used (Python: pyclustering.utils.metric).
  See: #626

• Corrected bug with Gower distance when numpy calculation is used and data shape is bigger than 1 (Python: pyclustering.utils.metric).
  See: #627

• Corrected MNDL splitting criteria for X-Means algorithm (Python: pyclustering.cluster.xmeans; C++: pyclustering::clst::xmeans).
  See: #623

pyclustering 0.10.0.1 release

pyclustering 0.10.0.1 library is a collection of clustering algorithms and methods, oscillatory networks, etc.

GENERAL CHANGES:

• Metadata of the library is updated.
  See: no reference

• Supported command test for setup.py script (Python: pyclustering).
  See: #607

• Introduced parameter random_seed for algorithms/models to control the seed of the random functionality: kmeans++, random_center_initializer, ga, gmeans, xmeans, som, somsc, elbow, silhouette_ksearch (Python: pyclustering.cluster; C++: pyclustering.clst).
  See: #578

• Introduced parameter k_max to G-Means algorithm to use it as an optional stop condition for the algorithm (Python: pyclustering.cluster.gmeans; C++: pyclustering::clst::gmeans).
  See: #602

• Implemented method save() for cluster_visualizer and cluster_visualizer_multidim to save visualization to file (Python: pyclustering.cluster).
  See: #601

• Optimization of CURE algorithm using balanced KD-tree (Python: pyclustering.cluster.cure; C++: pyclustering::clst::cure).
  See: #589

• Optimization of OPTICS algorithm using balanced KD-tree (Python: pyclustering.cluster.optics; C++: pyclustering::clst::optics).
  See: #588

• Optimization of DBSCAN algorithm using balanced KD-tree (Python: pyclustering.cluster.dbscan; C++: pyclustering::clst::dbscan).
  See: #587

• Implemented new optimized balanced KD-tree kdtree_balanced (Python: pyclustering.cluster.kdtree; C++: pyclustering::container::kdtree_balanced).
  See: #379

• Implemented KD-tree graphical visualizer kdtree_visualizer for KD-trees with 2-dimensional data (Python: pyclustering.container.kdtree).
  See: #586

• Updated interface of each clustering algorithm in C/C++ pyclustering cluster_data is substituted by concrete classes (C++ pyclustering::clst).
  See: #577

CORRECTED MAJOR BUGS:

• Bug with wrong data type for scores in Silhouette K-search algorithm in case of using C++ (Python: pyclustering.cluster.silhouette).
  See: #606

• Bug with a random distribution in the random center initializer (Python: pyclustering.cluster.center_initializer).
  See: #573

• Bug with incorrect converting Index List and Object List to Labeling when clusters do not contains one or more points from an input data (Python pyclustering.cluster.encoder).
  See: #596

• Bug with an exception in case of using user-defined metric for K-Means algorithm (Python pyclustering.cluster.kmeans).
  See: #600

• Memory leakage in the interface between python and C++ pyclustering library in case of CURE algorithm usage (C++ pyclustering).
  See: #581

pyclustering 0.10.0 pre-release

pyclustering 0.10.0 library is a collection of clustering algorithms and methods, oscillatory networks, etc.

GENERAL CHANGES:

• Supported command test for setup.py script (Python: pyclustering).
  See: #607

• Introduced parameter random_seed for algorithms/models to control the seed of the random functionality: kmeans++, random_center_initializer, ga, gmeans, xmeans, som, somsc, elbow, silhouette_ksearch (Python: pyclustering.cluster; C++: pyclustering.clst).
  See: #578

• Introduced parameter k_max to G-Means algorithm to use it as an optional stop condition for the algorithm (Python: pyclustering.cluster.gmeans; C++: pyclustering::clst::gmeans).
  See: #602

• Implemented method save() for cluster_visualizer and cluster_visualizer_multidim to save visualization to file (Python: pyclustering.cluster).
  See: #601

• Optimization of CURE algorithm using balanced KD-tree (Python: pyclustering.cluster.cure; C++: pyclustering::clst::cure).
  See: #589

• Optimization of OPTICS algorithm using balanced KD-tree (Python: pyclustering.cluster.optics; C++: pyclustering::clst::optics).
  See: #588

• Optimization of DBSCAN algorithm using balanced KD-tree (Python: pyclustering.cluster.dbscan; C++: pyclustering::clst::dbscan).
  See: #587

• Implemented new optimized balanced KD-tree kdtree_balanced (Python: pyclustering.cluster.kdtree; C++: pyclustering::container::kdtree_balanced).
  See: #379

• Implemented KD-tree graphical visualizer kdtree_visualizer for KD-trees with 2-dimensional data (Python: pyclustering.container.kdtree).
  See: #586

• Updated interface of each clustering algorithm in C/C++ pyclustering cluster_data is substituted by concrete classes (C++ pyclustering::clst).
  See: #577

CORRECTED MAJOR BUGS:

• Bug with wrong data type for scores in Silhouette K-search algorithm in case of using C++ (Python: pyclustering.cluster.silhouette).
  See: #606

• Bug with a random distribution in the random center initializer (Python: pyclustering.cluster.center_initializer).
  See: #573

• Bug with incorrect converting Index List and Object List to Labeling when clusters do not contains one or more points from an input data (Python pyclustering.cluster.encoder).
  See: #596

• Bug with an exception in case of using user-defined metric for K-Means algorithm (Python pyclustering.cluster.kmeans).
  See: #600

• Memory leakage in the interface between python and C++ pyclustering library in case of CURE algorithm usage (C++ pyclustering).
  See: #581

pyclustering 0.9.3.1 release

pyclustering 0.9.3.1 library is a collection of clustering algorithms and methods, oscillatory networks, etc.

CORRECTED MAJOR BUGS:

• Hotfix for the CF-tree - call method with incorrect amount of arguments.
  See: #570

pyclustering 0.9.3 release

pyclustering 0.9.3 library is a collection of clustering algorithms and methods, oscillatory networks, etc.

GENERAL CHANGES:

• Introduced get_cf_clusters and get_cf_entries methods for BIRCH algorithm to get CF-entry encoding information (pyclustering.cluster.birch).
  See: #569

• Introduced predict method for SOMSC algorithm to find closest clusters for specified points (pyclustering.cluster.somsc).
  See: #546

• Parallel optimization of C++ pyclustering compilation process.
  See: #553

• Include folder for easy integration to other C++ projects.
  See: #554

• Introduced new targets to build static libraries on Windows platform.
  See: #555

• Introduced new targets to build static libraries on Linux/MacOS platforms.
  See: #556

CORRECTED MAJOR BUGS:

• Bug with incorrect finding of closest CF-entry (pyclustering.container.cftree).
  See: #564

• Bug with incorrect BIRCH clustering due incorrect leaf analysis (pyclustering.cluster.birch).
  See: #563

• Bug with incorrect search procedure of farthest nodes in CF-tree (pyclustering.container.cftree).
  See: #551

• Bug with crash during clustering with the same points in case of BIRCH (pyclustering.cluster.birch).
  See: #561

pyclustering 0.9.2 release

pyclustering 0.9.2 library is a collection of clustering algorithms and methods, oscillatory networks, etc.

GENERAL CHANGES:

• Introduced checking of input arguments for clustering algorithm to provide human-readable errors (pyclustering.cluster).
  See: #548

• Implemented functionality to perform Anderson-Darling test for Gaussian distribution (ccore.stats).
  See: #550

• Implemented new clustering algorithm G-Means (pyclustering.cluster.gmeans, ccore.clst.gmeans).
  See: #506

• Introduced parameter repeat to improve parameters in X-Means algorithm (pyclustering.cluster.xmeans, ccore.clst.xmeans).
  See: #525

• Introduced new distance metric: Gower (pyclustering.utils.metric, ccore.utils.metric).
  See: #544

• Introduced sampling algorithms reservoir_r and reservoir_x (pyclustering.utils.sampling).
  See: #542

• Introduced parameter data_type to Silhouette method to use distance matrix (pyclustering.cluster.silhouette, ccore.clst.silhouette).
  See: #543

• Optimization of HHN (Hodgkin-Huxley Neural Network) by parallel processing (ccore.nnet.hhn).
  See: #541

• Introduced get_total_wce method for xmeans algorithm to find WCE (pyclustering.cluster.xmeans).
  See: #508

CORRECTED MAJOR BUGS:

• Incorrect center initialization in K-Means++ when candidates are not farthest (pyclustering.cluster.center_initializer).
  See: #549

pyclustering 0.9.1 release

pyclustering 0.9.1 library is a collection of clustering algorithms and methods, oscillatory networks, neural networks, etc.

GENERAL CHANGES:

• Introduced predict method for X-Means algorithm to find closest clusters for particular points (pyclustering.cluster.xmeans).
  See: #540

• Optimization of OPTICS algorithm by reducing complexity (ccore.clst.optics).
  See: #521

• Optimization of K-Medians algorithm by parallel processing (ccore.clst.kmedians).
  See: #529

• Introduced predict method for K-Medoids algorithm to find closest clusters for particular points (pyclustering.cluster.kmedoids).
  See: #527

• Introduced predict method for K-Means algorithm to find closest clusters for particular points (pyclustering.cluster.kmeans).
  See: #515

• Parallel optimization of Elbow method. (ccore.clst.elbow).
  See: #511

pyclustering 0.9.0 release

pyclustering 0.9.0 library is a collection of clustering algorithms and methods, oscillatory networks, neural networks, etc.

GENERAL CHANGES:

• CCORE (pyclustering core) is supported for MacOS.
  See: #486

• Introduced parallel Fuzzy C-Means algorithm (pyclustering.cluster.fcm, ccore.clst.fcm).
  See: #386

• Introduced new 'itermax' parameter for K-Means, K-Medians, K-Medoids algorithm to control maximum amount of iterations (pyclustering.cluster, ccore.clst).
  See: #496

• Implemented Silhouette and Silhouette K-Search algorithm for CCORE (ccore.clst.silhouette, ccore.clst.silhouette_ksearch).
  See: #490

• Implemented CLIQUE algorithms (pyclustering.cluster.clique, ccore.clst.clique).
  See: #381

• Introduced new distance metrics: Canberra and Chi Square (pyclustering.utils.metric, ccore.utils.metric).
  See: #482

• Optimization of CURE algorithm (C++ implementation) by using heap (multiset) instead of list to store clusters in queue (ccore.clst.cure).
  See: #479

CORRECTED MAJOR BUGS:

• Bug with crossover mask generation for genetic clustering algorithm (pyclustering.cluster.ga).
  See: #474

• Bug with hanging of K-Medians algorithm for some cases when algorithm is initialized by wrong amount of centers (ccore.clst.kmedians).
  See: #498

• Bug with incorrect center initialization, when the same point can be placed to result more than once (pyclustering.cluster.center_initializer, ccore.clst.kmeans_plus_plus).
  See: #497

• Bug with incorrect clustering in case of CURE python implementation when clusters are allocated incorrectly (pyclustering.cluster.cure).
  See: #483

• Bug with incorrect distance calculation for kmeans++ in case of index representation for centers (pyclustering.cluster.center_initializer).
  See: #485