Type:	Package
Title:	Detecting Class Overlapping Regions in Multidimensional Data
Version:	0.1.0
Description:	The issue of overlapping regions in multidimensional data arises when different classes or clusters share similar feature representations, making it challenging to delineate distinct boundaries between them accurately. This package provides methods for detecting and visualizing these overlapping regions using partitional clustering techniques based on nearest neighbor distances.
License:	GPL-3
Encoding:	UTF-8
Imports:	mclust, FNN, dplyr, stats, rlang
Suggests:	ggplot2
RoxygenNote:	7.3.1
URL:	https://github.com/pridiltal/clap
BugReports:	https://github.com/pridiltal/clap/issues
NeedsCompilation:	no
Packaged:	2024-06-28 10:25:59 UTC; User
Author:	Priyanga Dilini Talagala [aut, cre, cph]
Maintainer:	Priyanga Dilini Talagala <pritalagala@gmail.com>
Repository:	CRAN
Date/Publication:	2024-06-28 14:50:06 UTC

clap: Detecting Class Overlapping Regions in Multidimensional Dat

Description

The issue of overlapping regions in multidimensional data arises when different classes or clusters share similar feature representations, making it challenging to delineate distinct boundaries between them accurately. This package provides methods for detecting and visualizing these overlapping regions using partitional clustering techniques based on nearest neighbor distances.

Author(s)

Priyanga Dilini Talagala _PACKAGE

Compute cluster composition and filter based on percentage

Description

This function computes the cluster composition based on the input object of class 'clap' returned by perform_clustering function. It merges the data and cluster assignments, computes cluster composition statistics including counts, IDs, and percentages, and filters based on the specified percentage threshold.

Usage

compute_cluster_composition(x)

Arguments

Value

filtered data frame summarizing cluster composition with class 'clap'.

Examples

if (requireNamespace("ggplot2", quietly = TRUE)) {
  # Generate dummy data
  class1 <- matrix(rnorm(100, mean = 0, sd = 1), ncol = 2) +
    matrix(rep(c(1, 1), each = 50), ncol = 2)
  class2 <- matrix(rnorm(100, mean = 0, sd = 1), ncol = 2) +
    matrix(rep(c(-1, -1), each = 50), ncol = 2)
  datanew <- rbind(class1, class2)
  training <- data.frame(datanew, class = factor(c(rep(1, 50), rep(2, 50))))

  # Plot the dummy data to visualize overlaps
  p <- ggplot2::ggplot(training, ggplot2::aes(x = X1, y = X2, color = class)) +
    ggplot2::geom_point() +
    ggplot2::labs(title = "Dummy Data with Overlapping Classes")
  print(p)

  # Perform clustering
  cluster_result <- perform_clustering(training, class_column = class)
  # Compute cluster composition
  composition <- compute_cluster_composition(cluster_result)
}

Extract and convert IDs to numeric vector

Description

This function extracts IDs from a data frame containing filtered composition data and converts them into a numeric vector.

Usage

extract_ids_vector(composition)

Arguments

Value

A numeric vector of IDs.

Examples

if (requireNamespace("ggplot2", quietly = TRUE)) {
  # Generate dummy data
  class1 <- matrix(rnorm(100, mean = 0, sd = 1), ncol = 2) +
    matrix(rep(c(1, 1), each = 50), ncol = 2)
  class2 <- matrix(rnorm(100, mean = 0, sd = 1), ncol = 2) +
    matrix(rep(c(-1, -1), each = 50), ncol = 2)
  datanew <- rbind(class1, class2)
  training <- data.frame(datanew, class = factor(c(rep(1, 50), rep(2, 50))))

  # Plot the dummy data to visualize overlaps
  p <- ggplot2::ggplot(training, ggplot2::aes(x = X1, y = X2, color = class)) +
    ggplot2::geom_point() +
    ggplot2::labs(title = "Dummy Data with Overlapping Classes")
  print(p)

  # Perform clustering
  cluster_result <- perform_clustering(training, class_column = class)
  # Compute cluster composition
  composition <- compute_cluster_composition(cluster_result)
  # Extract IDs to numeric vector
  ids_vector <- extract_ids_vector(composition)
  # Subset data based on extracted IDs
  overlapdata <- training[ids_vector, ]
  # Plot overlapping data points
  p2 <- p + ggplot2::geom_point(data = overlapdata, ggplot2::aes(X1, X2), colour = "black")
  print(p2)
}

Perform clustering based on nearest neighbor distances

Description

Perform clustering based on nearest neighbor distances

Usage

perform_clustering(data, class_column = NULL)

Arguments

Details

This function first removes the specified class column from the data, calculates the nearest neighbor distances, and then performs clustering using a radius based on the maximum nearest neighbor distance.

Value

An object of class 'clap' containing:

members

A list of clusters with their respective data point IDs.

cluster_df

A data frame with cluster assignments for each data point.

data

The original dataset.

Examples

if (requireNamespace("ggplot2", quietly = TRUE)) {
  # Generate dummy data
  class1 <- matrix(rnorm(100, mean = 0, sd = 1), ncol = 2) +
    matrix(rep(c(1, 1), each = 50), ncol = 2)
  class2 <- matrix(rnorm(100, mean = 0, sd = 1), ncol = 2) +
    matrix(rep(c(-1, -1), each = 50), ncol = 2)
  datanew <- rbind(class1, class2)
  training <- data.frame(datanew, class = factor(c(rep(1, 50), rep(2, 50))))

  # Plot the dummy data to visualize overlaps
  p <- ggplot2::ggplot(training, ggplot2::aes(x = X1, y = X2, color = class)) +
    ggplot2::geom_point() +
    ggplot2::labs(title = "Dummy Data with Overlapping Classes")
  print(p)

  # Perform clustering
  cluster_result <- perform_clustering(training, class_column = class)
}