2025-12-12
2025-12-08
2025-12-02
2025-11-28
2025-11-04
nthread) updated. 2025-11-07
2025-10-23
2025-8-21
2025-2-06
2025-2-04
2024-7-20
2024-4-23
2024-4-15
2024-3-16
2023-10-03
2023-09-02
2022-09-13
2022-08-27
2022-08-25
2022-08-20
2022-08-17
[IMPROVEMENT] Renamed the grid_method “Dp compressed likelihood” to “DP exact likelihood”, “Dp compressed likelihood nlog” to “Dp approx likelihood”.
[IMPROVEMENT] Removed ULONG_MAX in “Cut_by_density.cpp” to fix compiling issues under Linux. Similarly, DBL_MAX and INT_MIN were removed in Cutting_Cluster_dp_compressed.cpp.
[BUG FIX] “#include
[BUG FIX] Removed the usage of “cref()” in “Cutting_Cluster_dp.cpp” and “Cutting_Cluster_dp_compressed.cpp” to avoid compiling error on Linux.
[NEW FEATURE] Adding test files for “DP Compressed likelihood nlogn” and “DP Compressed likelihood”in testthat.
[NEW FEATURE] Added the perimeter “noise” for user to determined whether adding jitter noise to original data or not. The default value is FALSE which no noise will be added. Only will be used when cluster_method = “BALL+BIC”.
[MAJOR NEW FEATURE] Added a new grid method “DP Compressed likelihood nlogn”, which is a speed up version of “DP Compressed likelihood” by using Divide and Conquer when filling the matrix, but no optimal will be generated. The “DP Compressed likelihood nlogn” is set to default for the parameter “grid_method”.
[IMPROVEMENT] Optimized the structure of the cluster class in c code to speed up and make the code more robust, and rewrite the “Sort+Split” part correspondingly.
[NEW FEATURE] Added cluster_method and grid_method to the result, if the cluster label is given by user, the cluster_method will be set to “user supply”.
[IMPROVEMENT] Modify the plot function to print both methods in title.
2022-08-07
[BUG FIX] Replaced ULONG_MAX by
std::numeric_limits
[BUG FIX] Included header file
[BUG FIX] “#include
2022-07-27
2022-06-30
Version 0.1.3 Created from version 0.1.2.
[NEW FEATURE] Used dynamic programming to maximize the likelihood of categorical distribution, to improve quality of discretization
[NEW FEATURE] Used BIC on the categorical likelihood to select number of discrete levels for each variable.
[IMPROVEMENT] Compression: avoid trying discretization within consecutive points of the same cluster label to greatly reduce runtime of dynamic programming.
[BUG FIX] The Examples.Rmd vignette does not compile on MacOS.
[BUG FIX] Five test cases failed.
2022-03-17
2022-03-10
[NEW FEATURE] Added new cluster methods “DP” and “DP Compressed”. The new method using dynamic programming to find the optimal solution to split the label into many zone. The maximum number of zone is set to 2*K by hand currently.
[NEW FEATURE] Implemented a new approach of the “DP” method. Improved by compressed the label, the size of the dynamic programming table can be decreased dramatically.
[IMPROVEMENT] Before returning the bins on each dimension, we will check the upsilon static to make sure the cutting is reasonable. Currently the cut off has been set to the median of the PDF. Any result with a upsilon static that smaller than the median of the PDF will be drop and no cut will be put on that dimension.
2022-01-25
2022-01-18
2022-01-17
2022-01-11
2021-10-26
2021-10-06
2020-09-13
2020-08-10
2020-04-03
2020-03-31
2020-03-26
2020-03-25
2020-03-24
2020-03-20
2020-03-17
2020-03-14
2020-03-09