Creacion de datos para clusters
Freddy Hernandez
10 octubre, 2019
1 Paquete
El paquete necesario para crear datos usados en clustering es clusteringdatasets y la instalación se puede hacer usando el siguiente código.
if (!require('devtools')) install.packages('devtools')
devtools::install_github('elbamos/clusteringdatasets', force=TRUE)Cargando el paquete.
library(clusteringdatasets)2 Gotas o Blobs
Para hacer gotas o blobs se usa la siguiente función.
make_blobs(n_samples = 100, n_features = 2, centers = 3,
cluster_std = 1, center_box = c(-10, 10), shuffle = TRUE)2.1 Ejemplo
Crear 100 observaciones de 3 grupos cada uno con dos variables y usando los vectores de medias.
medias <- matrix(c(0, 0,
3, 1,
2, 3), ncol=2, byrow=TRUE)
blobs <- make_blobs(n_samples=1000, n_features=2, centers=medias)
plot(blobs$samples, col=rainbow(3)[blobs$labels])
2.2 Ejemplo
Crear 10 observaciones de 3 grupos cada uno con 3 variables y usando los vectores de medias.
medias <- matrix(c(0, 0, 0,
1, 1, 0,
1, 1, 1), ncol=3, byrow=TRUE)
blobs <- make_blobs(n_samples=10, n_features=3, centers=medias)
blobs## $samples
## [,1] [,2] [,3]
## [1,] -0.4921498 -0.41712455 1.50333632
## [2,] 0.8606919 1.64510869 1.11452831
## [3,] -0.2404758 1.43739233 -0.57638507
## [4,] -0.6796860 -1.62739298 0.06563251
## [5,] 1.5592125 2.44386438 2.39827294
## [6,] -0.2938455 2.28756286 -0.55289562
## [7,] 1.9123192 1.47654203 -0.55703259
## [8,] 0.0959438 2.45646794 1.07661910
## [9,] -1.0876149 -1.70068738 0.80432385
## [10,] 0.2735340 0.06982408 0.33965061
##
## $labels
## [1] 1 2 2 1 3 2 1 3 1 33 Lunas o Moons
Para hacer lunas se usa la siguiente función.
make_moons(n_samples = 100, shuffle = TRUE, noise = NA)3.1 Ejemplo
Crear 1000 observaciones de dos grupos con formas de luna.
moons <- make_moons(n_samples=1000, shuffle=FALSE, noise=0.15)
plot(moons$samples, col=rainbow(2)[moons$labels])
4 Birch
data(birch1)
data(birch2)
data(birch3)
par(mfrow = c(1, 3), mar = c(0,0,1,0))
plot(birch1, cex = 0.0000005, main = "birch1", xlab = "", ylab = NULL, xaxt='n', yaxt = 'n')
plot(birch2, cex = 0.0000005, main = "birch2", xlab = NULL, ylab = NULL, xaxt='n',yaxt = 'n')
plot(birch3, cex = 0.0000005, main = "birch3", xlab = NULL, ylab = NULL, xaxt='n',yaxt = 'n')
5 S Sets
The S-sets are useful for testing how an algorithm handles cluster overlap.
data(s1)
data(s2)
data(s3)
data(s4)
par(mfrow = c(2, 2), mar = c(0,0,1,0))
plot(s1[, 1:2], cex = 0.0001, col = s1$labels, main = "s1", xlab = NULL, ylab = NULL, xaxt='n',yaxt = 'n')
plot(s2[, 1:2], cex = 0.0001, col = s2$labels, main = "s2", xlab = NULL, ylab = NULL, xaxt='n',yaxt = 'n')
plot(s3[, 1:2], cex = 0.0001, col = s3$labels, main = "s3", xlab = NULL, ylab = NULL, xaxt='n',yaxt = 'n')
plot(s4[, 1:2], cex = 0.0001, col = s4$labels, main = "s4", xlab = NULL, ylab = NULL, xaxt='n',yaxt = 'n')
6 A Sets
data(a1)
data(a2)
data(a3)
par(mfrow = c(1, 3), mar = c(0,0,1,0))
plot(a1[, 1:2], cex = 0.0001, col = s1$labels, main = "a1", xlab = NULL, ylab = NULL, xaxt='n',yaxt = 'n')
plot(a2[, 1:2], cex = 0.0001, col = s2$labels, main = "a2", xlab = NULL, ylab = NULL, xaxt='n',yaxt = 'n')
plot(a3[, 1:2], cex = 0.0001, col = s3$labels, main = "a3", xlab = NULL, ylab = NULL, xaxt='n',yaxt = 'n')
7 Shapesets
data("Aggregation")
data("spiral")
data("D31")
data(Compound)
data(pathbased)
data(jain)
data(flame)
data(R15)
par(mfrow = c(3, 3), mar = c(0,0,1,0))
plot(Aggregation[, 1:2], cex = 0.1, col = Aggregation$label, main = "Aggregation", xlab = NULL, ylab = NULL, xaxt='n',yaxt = 'n')
plot(spiral[, 1:2], cex = 0.1, col = spiral$label, main = "spiral", xlab = NULL, ylab = NULL, xaxt='n',yaxt = 'n')
plot(D31[, 1:2], cex = 0.1, col = D31$label, main = "D31", xlab = NULL, ylab = NULL, xaxt='n',yaxt = 'n')
plot(Compound[, 1:2], cex = 0.1, col = Compound$label, main = "Compound", xlab = NULL, ylab = NULL, xaxt='n',yaxt = 'n')
plot(pathbased[, 1:2], cex = 0.1, col = pathbased$label, main = "pathbased", xlab = NULL, ylab = NULL, xaxt='n',yaxt = 'n')
plot(jain[, 1:2], cex = 0.1, col = jain$label, main = "jain", xlab = NULL, ylab = NULL, xaxt='n',yaxt = 'n')
plot(flame[, 1:2], cex = 0.1, col = flame$label, main = "flame", xlab = NULL, ylab = NULL, xaxt='n',yaxt = 'n')
plot(R15[, 1:2], cex = 0.1, col = R15$label, main = "R15", xlab = NULL, ylab = NULL, xaxt='n',yaxt = 'n')
8 Chameleon
par(mfrow = c(2, 2), mar = c(0,0,1,0))
data("t48k")
data("t58k")
data("t710k")
data("t88k")
plot(t48k, cex = 0.001, main = "t48k", xlab = NULL, ylab = NULL, xaxt='n',yaxt = 'n')
plot(t58k, cex = 0.001, main = "t58k", xlab = NULL, ylab = NULL, xaxt='n',yaxt = 'n')
plot(t710k, cex = 0.001, main = "t710k", xlab = NULL, ylab = NULL, xaxt='n',yaxt = 'n')
plot(t88k, cex = 0.001, main = "t88k", xlab = NULL, ylab = NULL, xaxt='n',yaxt = 'n')
9 Neural Gas
names <- c("Circle" , "Complex1" , "Complex2" , "Complex3" , "Complex4" , "Discrete" , "HiLoDensity" , "JumpingRectangle" ,
"MovingJumpingRectangle", "MovingRectangle", "Rectangle" , "RMouseRectangle" ) # "Ring"
data(list = names)
par(mfrow = c(3, 5), mar = c(0, 0, 1, 0))
for (nm in names) {
plot(eval(parse(text = nm)), cex = 0.01, main = nm, xlab = NULL, ylab = NULL, xaxt='n',yaxt = 'n')
}
10 Non-Convex
names <- c("cross" , "d4" , "face" , "pie" , "ring2" , "sincos")
data(list = names)
par(mfrow = c(2, 3), mar = c(0, 0, 1, 0))
for (nm in names) {
plot(eval(parse(text = nm)), cex = 0.1, main = nm, xlab = NULL, ylab = NULL, xaxt='n',yaxt = 'n')
}
11 Locations
data("mopsifinland")
data("mopsijoensu")
par(mfrow = c(1, 2), mar = c(0,0,1,0))
plot(mopsifinland[, 1:2], cex = 0.01, main = "mopsifinland", xlab = NULL, ylab = NULL, xaxt='n',yaxt = 'n')
plot(mopsijoensu[, 1:2], cex = 0.05, main = "mopsifinland", xlab = NULL, ylab = NULL, xaxt='n',yaxt = 'n')
12 High Dimensional Datasets
The package contains three sets of high-dimensional data. The visualizations below were made using my largeVis package to reduce each dataset to two dimensions, and the colors are the result of applying the hdbscan function within the package.
12.0.1 UCI Datasets
library(largeVis)
library(clusteringdatasets)
library(ggplot2)
data(glass)
data(wdbc)
data(breast)
data(yeast)
data(wine)
data(thyroid)
toproc <- list(glass, wdbc, breast, yeast, wine, thyroid)
vises <- list()
clusters <- list()
for (i in 1:length(toproc)) {
dat <- t(scale(as.matrix(toproc[[i]])))
if (ncol(dat) < 50000) vis <- largeVis(dat, K = 50, verbose = TRUE)
else vis <- largeVis(dat, K = 100, verbose = TRUE)
neighbors <- randomProjectionTreeSearch(dat, K = 50)
edges <- buildEdgeMatrix(data = dat, neighbors = neighbors)
print(str(edges))
cluster <- hdbscan(edges = edges,neighbors = neighbors, K = 5, minPts = 10, verbose = TRUE)
vises[[i]] <- vis
clusters[[i]] <- cluster
}library(ggplot2)
load(system.file("extdata/vises.Rda", package = "clusteringdatasets"))
load(system.file("extdata/clusters.Rda", package = "clusteringdatasets"))
names <- c("glass", "wdbc", "breast", "yeast", "wine", "thyroid")
par(mfrow = c(2, 3), mar = c(0,0,1,0))
for (i in 1:length(names)) {
df <- data.frame(t(vises[[i]]$coords))
colnames(df) <- c("x", "y")
df$label <- clusters[[i]]$clusters
if (length(unique(df$label)) > 1) {
plot(df[, 1:2], cex = 0.005, col = df$label, main = names[i], xlab = NULL, ylab = NULL, xaxt='n',yaxt = 'n')
} else {
plot(df[, 1:2], cex = 0.005, col = df$label, main = names[i], xlab = NULL, ylab = NULL, xaxt='n',yaxt = 'n')
}
}
13 KDDCUP04Bio
data("kddcup04bio")
library(largeVis)
load("./kddvis.Rda")
dat <- t(scale(as.matrix(kddcup04bio)))
vis <- largeVis(dat, K = 50, n_trees = 50, tree_threshold = 50, max_iter = 2, verbose = TRUE)load(system.file("extdata/kdvis.Rda", package = "clusteringdatasets"))
par(mfrow = c(1, 1), mar = c(0,0,1,0))
plot(kdvis[, 1:2], cex = 0.0001, col = df$label, main = "kddcup04bio", xlab = NULL, ylab = NULL, xaxt='n',yaxt = 'n',
xlim = c(-20, 18), ylim = c(-20, 30))