| Type: | Package |
| Depends: | R (≥ 3.5.0) |
| Title: | Whale Optimization Algorithm for K-Medoids Clustering |
| Version: | 0.2.2 |
| Date: | 2026-02-08 |
| Encoding: | UTF-8 |
| Description: | Implements the Whale Optimization Algorithm(WOA) for k-medoids clustering, providing tools for effective and efficient cluster analysis in various data sets. The methodology is based on "The Whale Optimization Algorithm" by Mirjalili and Lewis (2016) <doi:10.1016/j.advengsoft.2016.01.008>. |
| License: | GPL-2 | GPL-3 [expanded from: GPL (≥ 2)] |
| Imports: | dtwclust, proxy, cluster, Rcpp (≥ 1.0.11), RcppParallel |
| LinkingTo: | Rcpp, RcppParallel |
| SystemRequirements: | GNU make |
| RoxygenNote: | 7.3.3 |
| LazyData: | true |
| NeedsCompilation: | yes |
| Packaged: | 2026-02-18 13:56:36 UTC; huang |
| Author: | Chenan Huang [aut, cre],
Narumasa Tsutsumida
 [aut] |
| Maintainer: | Chenan Huang <hualianchan@gmail.com> |
| Repository: | CRAN |
| Date/Publication: | 2026-02-18 14:20:09 UTC |
Lightning7 Data for Testing
Description
A dataset containing example data for testing purposes from the UCR Time Series Classification Archive. This dataset is a time series dataset with correct classifications in the first column. There are 7 classes in this dataset. It contains 73 series, each with 319 time points, and the best DTW window length for this dataset is 5.
Usage
data(Lightning7)
Format
A data frame with 73 rows and 320 columns. The first column (V1) is a factor vector of correct classifications, and the remaining 319 columns (V2 to V320) are numeric vectors of time series data.
Source
UCR Time Series Classification Archive
References
Eads, Damian R., et al. "Genetic algorithms and support vector machines for time series classification." Applications and Science of Neural Networks, Fuzzy Systems, and Evolutionary Computation V. Vol. 4787. International Society for Optics and Photonics, 2002.
-
http://www.timeseriesclassification.com/description.php?Dataset=Lightning7
Examples
data(Lightning7)
head(Lightning7)
Fitness function for WOA optimization (C++ implementation)
Description
Combines cluster assignment and total distance calculation. Returns Inf if any cluster has fewer than 2 members.
Usage
fitnessFunction_cpp(dist_vec, nrows, medoids)
Arguments
dist_vec |
Distance vector (from dist object) |
nrows |
Number of rows in the original data |
medoids |
Integer vector of medoid indices (1-indexed) |
Value
total_dist (double), Inf if invalid clustering
Deterministic serial fitness function for final evaluation
Description
Deterministic serial fitness function for final evaluation
Usage
fitnessFunction_serial_cpp(dist_vec, nrows, medoids)
Arguments
dist_vec |
Distance vector (from dist object) |
nrows |
Number of rows in the original data |
medoids |
Integer vector of medoid indices (1-indexed) |
Value
total_dist (double), Inf if invalid clustering
Check if two medoid solutions are identical
Description
Check if two medoid solutions are identical
Usage
medoidsEqual_cpp(medoids1, medoids2)
Arguments
medoids1 |
First medoid vector (1-indexed) |
medoids2 |
Second medoid vector (1-indexed) |
Value
TRUE if identical (regardless of order), FALSE otherwise
Project MDS coordinates to nearest unique sample indices
Description
Finds the nearest sample in the MDS embedded space for each medoid coordinate, while enforcing a one-to-one mapping between medoids and samples (no duplicates).
Usage
projectToIndex_cpp(medoid_coords, Z)
Arguments
medoid_coords |
NumericMatrix of medoid coordinates in MDS space (ClusNum x mds_dim) |
Z |
NumericMatrix of sample coordinates in MDS space (nrows x mds_dim) |
Value
IntegerVector of nearest unique sample indices (1-indexed, length = ClusNum)
Whale Optimization Algorithm for K-Medoids Clustering
Description
This function implements the Whale Optimization Algorithm (WOA) for K-Medoids clustering. Supported distance measures are Dynamic Time Warping (DTW) and Euclidean Distance (ED).
Usage
woa_kmedoids(
data,
ClusNum,
distance_method = c("dtw", "ed"),
learned_w = NULL,
Max_iter = 200,
n = 5,
early_stopping = TRUE,
patience = 5,
verbose = FALSE
)
Arguments
data |
Data matrix |
ClusNum |
Number of clusters |
distance_method |
Distance calculation method, either "dtw" or "ed" |
learned_w |
Window size for DTW (only used if distance_method is "dtw") |
Max_iter |
Maximum number of iterations (default is 200, it can be adjusted according to the size of the dataset) |
n |
Population size (number of whales, default is 5, it can be adjusted according to the size of the dataset) |
early_stopping |
Logical. If TRUE, stop early when the best solution converges (default is TRUE) |
patience |
Number of consecutive iterations without improvement before early stopping (default is 5) |
verbose |
Logical. If TRUE, print progress messages (default is FALSE) |
Value
The 'woa_clustering' object containing the clustering result and medoids
Author(s)
Chenan Huang, Narumasa Tsutsumida
References
Chenan H. and Tsutsumida N. (2025) A scalable k-medoids clustering via whale optimization algorithm, Array, 28,100599. https://doi.org/10.1016/j.array.2025.100599.
Examples
# NOTE: This example only shows how to implement woa_kmedoids using sample data.
# Results do not suggest any meanings.
data(Lightning7)
Lightning7_data <- Lightning7[, -1] # Remove the first column of classification data
result <- woa_kmedoids(Lightning7_data, ClusNum = 7, distance_method = "dtw", learned_w = 5)
print(result)