FuzzyDBScan (version 0.0.3)

Fuzzy_DBScan: Fuzzy DBScan

Description

This object implements fuzzy DBScan with both, fuzzy cores and fuzzy borders. Additionally, it provides a predict function.

Public fields

dta

data.frame | matrix
The data to be clustered by the algorithm. Allowed are only numeric columns.

eps

numeric
The size (radius) of the epsilon neighborhood. If the radius contains 2 numbers, the fuzzy cores are calculated between the minimum and the maximum radius. If epsilon is a single number, the algorithm looses the fuzzy core property. If the length of pts is also 1L, the algorithm equals to non-fuzzy DBScan.

pts

numeric
number of maximum and minimum points required in the eps neighborhood for core points (excluding the point itself). If the length of the argument is 1, the algorithm looses its fuzzy border property. If the length of eps is also 1L, the algorithm equals to non-fuzzy DBScan.

clusters

factor
Contains the assigned clusters per observation in the same order as in dta.

dense

numeric
Contains the assigned density estimates per observation in the same order as in dta.

point_def

character
Contains the assigned definition estimates per observation in the same order as in dta. Possible are "Core Point", "Border Point" and "Noise".

results

data.table
A table where each column indicates for the probability of the new data to belong to a respective cluster.

Methods

Public methods

• FuzzyDBScan$new()

• FuzzyDBScan$predict()

• FuzzyDBScan$plot()

• FuzzyDBScan$clone()

---

Method new()

Create a FuzzyDBScan object. Apply the fuzzy DBScan algorithm given the data dta, the range of the radius eps and the range of the Points pts.

Usage

FuzzyDBScan$new(dta, eps, pts)

Arguments

dta

data.frame | matrix
The data to be clustered by the algorithm. Allowed are only numeric columns.

eps

numeric
The size (radius) of the epsilon neighborhood. If the radius contains 2 numbers, the fuzzy cores are calculated between the minimum and the maximum radius. If epsilon is a single number, the algorithm looses the fuzzy core property. If the length of pts is also 1L, the algorithm equals to non-fuzzy DBScan.

pts

numeric
number of maximum and minimum points required in the eps neighborhood for core points (excluding the point itself). If the length of the argument is 1, the algorithm looses its fuzzy border property. If the length of eps is also 1L, the algorithm equals to non-fuzzy DBScan.

---

Method predict()

Predict new data with the initialized algorithm.

Usage

FuzzyDBScan$predict(new_data, cmatrix = TRUE)

Arguments

new_data

data.frame | matrix
The data to be predicted by the algorithm. Allowed are only numeric columns which should match to self$dta.

cmatrix

logical
Indicating whether the assigned cluster should be returned in form of a matrix where each column indicates for the probability of the new data to belong to a respective cluster. The object will have the same shape as the results field. If set to FALSE the shape of the returned assigned clusters is a two-column data.table with one column indicating the assigned cluster and the second column indicating the respective probability of the new data.

---

Method plot()

Plot clusters and soft labels on two features.

Usage

FuzzyDBScan$plot(x, y)

Arguments

x

character
Feature to plot on the x-axis.

y

character
Feature to plot on the y-axis.

---

Method clone()

The objects of this class are cloneable with this method.

Usage

FuzzyDBScan$clone(deep = FALSE)

Arguments

deep

Whether to make a deep clone.

Details

A method to initialize and run the algorithm and a function to predict new data. The package is build upon the paper "Fuzzy Extensions of the DBScan algorithm" from Ienco and Bordogna. The predict function assigns new data based on the same criteria as the algorithm itself. However, the prediction function freezes the algorithm to preserve the trained cluster structure and treats each new prediction object individually. Note, that border points are included to the cluster.

References

Ienco, Dino, and Gloria Bordogna. Fuzzy extensions of the DBScan clustering algorithm. Soft Computing 22.5 (2018): 1719-1730.

Examples

# load factoextra for data and ggplot for plotting
library(factoextra)
dta = multishapes[, 1:2]
eps = c(0, 0.2)
pts = c(3, 15)
# train DBScan based on data, ep and pts
cl = FuzzyDBScan$new(dta, eps, pts)
# Plot DBScan for x and y
library(ggplot2)
cl$plot("x", "y")
# produce test data
x <- seq(min(dta$x), max(dta$x), length.out = 50)
y <- seq(min(dta$y), max(dta$y), length.out = 50)
p_dta = expand.grid(x = x, y = y)
# predict on test data and plot results
p = cl$predict(p_dta, FALSE)
ggplot(p, aes(x = p_dta[, 1], y = p_dta[, 2], colour = as.factor(cluster))) +
  geom_point(alpha = p$dense)