-
Notifications
You must be signed in to change notification settings - Fork 62
/
Copy pathslides_clustering.Rmd
259 lines (194 loc) · 6.36 KB
/
slides_clustering.Rmd
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
---
title: "Clustering"
author: "Dr. Stephen W. Thomas, Queen's University"
output:
pdf_document:
highlight: pygments
number_sections: yes
toc: no
toc_depth: '2'
---
```{r}
library(dplyr)
library(ggplot2)
library(readr)
library(extrafont)
library(ggthemes)
library(caret)
library(cluster) # For preprocessing
library(factoextra) # for elbow plots
library(amap) # For distance metrics
library(stylo) # For cosine distance
loadfonts(quiet = T)
```
# Load Data
```{r}
dat <- read_csv("data/Mall_Customers_With_output.csv")
str(dat)
dat %>%
select(CustomerID, Genre, Age, AnnualIncome, SpendingScore) %>%
head(n=15)
```
```{r}
plotClust = function(df, colorCol) {
ggplot(df, aes(x=AnnualIncome, y=SpendingScore)) +
geom_point(aes(color=factor(colorCol)), size=6) +
geom_rangeframe() +
theme_tufte(base_family="Calibri") +
#theme(legend.position="none") +
theme(axis.title.x = element_text(colour="grey20",size=30,face="plain")) +
theme(axis.title.y = element_text(colour="grey20",size=30,face="plain")) +
theme(axis.text.x = element_text(colour="grey20",size=20,face="plain")) +
theme(axis.text.y = element_text(colour="grey20",size=20,face="plain")) +
labs(x = "Annual Income (K)", y ="Spending Score")
}
```
# Plot the data
```{r}
iwidth = 10
iheight = 7
theme_set(theme_gray(base_size = 32))
ggplot(dat, aes(x=AnnualIncome, y=SpendingScore)) +
geom_point(color="black", size=10) +
geom_rangeframe() +
theme_tufte(base_family="Calibri") +
theme(legend.position="none") +
theme(axis.title.x = element_text(colour="grey20",size=40,face="plain")) +
theme(axis.title.y = element_text(colour="grey20",size=40,face="plain")) +
theme(axis.text.x = element_text(colour="grey20",size=32,face="plain")) +
theme(axis.text.y = element_text(colour="grey20",size=32,face="plain")) +
labs(x = "Annual Income (K)", y ="Spending Score")
#ggsave(file="out/cluster_1.png", width=iwidth, height=iheight)
```
# Scaling the data
```{r}
pre = preProcess(dat[,c("Age", "AnnualIncome", "SpendingScore")], method="range")
dat.scaled = predict(pre, dat, method=c("range"))
head(dat)
head(dat.scaled)
dat.scaled %>%
select(CustomerID, Genre, Age, AnnualIncome, SpendingScore) %>%
head(n=15)
```
# Playing with Distance Metrics
```{r}
df = dat.scaled[,c("Age", "AnnualIncome", "SpendingScore")]
dd_euc = as.matrix(Dist(df, method = "euclidean"))
dd_man = as.matrix(Dist(df, method = "manhattan"))
dd_cos = as.matrix(dist.cosine(as.matrix(df)))
# Check ou the three closest and farthest people from Customer ID 1
print("euc")
n = ncol(dd_euc)
idx = order(dd_euc[1,])[2:4]
idx
dd_euc[1, idx]
df[c(1, idx),]
idx_long = order(dd_euc[1,])[(n-2):n]
idx_long
dd_euc[1, idx_long]
df[c(1, idx_long),]
print("man")
n = ncol(dd_man)
idx = order(dd_man[1,])[2:4]
idx
dd_man[1, idx]
df[c(1, idx),]
idx_long = order(dd_man[1,])[(n-2):n]
idx_long
dd_man[1, idx_long]
df[c(1, idx_long),]
print("cos")
n = ncol(dd_cos)
idx = order(dd_cos[1,])[2:4]
idx
dd_cos[1, idx]
df[c(1, idx),]
idx_long = order(dd_cos[1,])[(n-2):n]
idx_long
dd_cos[1, idx_long]
df[c(1, idx_long),]
```
# K-means
```{r}
set.seed(123)
clusters = kmeans(dat.scaled[,c("AnnualIncome", "SpendingScore")], centers=5, nstart=10)
dat.scaled$kmeans5 = clusters$cluster
clusters$tot.withinss
plotClust(dat.scaled, clusters$cluster)
ggsave(file="out/cluster_kmeans5.png", width=iwidth, height=iheight)
head(dat.scaled, n=15)
clusters$centers
clusters$size
clusters = kmeans(dat.scaled[,c("AnnualIncome", "SpendingScore")],
centers=3,
nstart=10)
dat.scaled$kmeans3 = clusters$cluster
clusters$tot.withinss
head(dat.scaled, n=15)
plotClust(dat.scaled, clusters$cluster)
ggsave(file="out/cluster_kmeans3.png", width=iwidth, height=iheight)
clusters = kmeans(dat.scaled[,c("AnnualIncome", "SpendingScore")], centers=10, nstart=10)
plotClust(dat, clusters$cluster)
#ggsave(file="out/cluster_kmeans1.png", width=iwidth, height=iheight)
```
## Elbow Plot
```{r}
set.seed(123)
fviz_nbclust(dat.scaled[,c("AnnualIncome", "SpendingScore")], kmeans, method = "silhouette", k.max=15)
ggsave(file="out/kmeans_sil.png", width=6, height=4)
fviz_nbclust(dat.scaled[,c("AnnualIncome", "SpendingScore")], kmeans, method = "wss", k.max=15)
ggsave(file="out/kmeans_wss.png", width=6, height=4)
```
## Which Distance Measure?
```{r}
df = dat.scaled[,c("AnnualIncome", "SpendingScore")]
dd_euc = dist(df, method = "euclidean")
dd_man = dist(df, method = "manhattan")
dd_min = dist(df, method = "minkowski")
k <- 5
set.seed(123)
res = data.frame(method=character(), wss=numeric(), betweenss=numeric())
clusters = kmeans(dd_euc, k, nstart=50)
res = rbind(res, data.frame(method="euclidean", wss=clusters$tot.withinss, betweenness=clusters$betweenss))
plotClust(df, clusters$cluster)
ggsave(file="out/kmeans_euc.png", width=iwidth, height=iheight)
clusters = kmeans(dd_man, k, nstart=50)
res = rbind(res, data.frame(method="manhattan", wss=clusters$tot.withinss, betweenness=clusters$betweenss))
plotClust(df, clusters$cluster)
ggsave(file="out/kmeans_man.png", width=iwidth, height=iheight)
clusters = kmeans(dd_min, k, nstart=50)
res = rbind(res, data.frame(method="minkowski", wss=clusters$tot.withinss, betweenness=clusters$betweenss))
plotClust(df, clusters$cluster)
ggsave(file="out/kmeans_min.png", width=iwidth, height=iheight)
res
```
# DBSCAN
```{r}
library(dbscan)
set.seed(123)
kNNdistplot(dat.scaled[,c("AnnualIncome", "SpendingScore")], k = 3)
# Right Click on the image below, then "Copy"
clusters = dbscan(dat.scaled[,c("Age", "AnnualIncome", "SpendingScore")],
minPts = 3,
eps = .14)
dat.scaled$dbscan = clusters$cluster
plotClust(dat.scaled, dat.scaled$dbscan)
ggsave(file="out/dbscan.png", width=iwidth, height=iheight)
```
# Hierarchical
```{r}
set.seed(123)
clusters = hclust(dist(dat.scaled[,c("AnnualIncome", "SpendingScore")]))
hcd <- as.dendrogram(clusters)
plot(hcd)
plot(hcd, xlim = c(1, 20), ylim = c(0,0.4))
dat.scaled$hclust5 = cutree(clusters, 5)
plotClust(dat.scaled, dat.scaled$hclust5)
ggsave(file="out/cluster_hclust5.png", width=iwidth, height=iheight)
dat.scaled$hclust3 = cutree(clusters, 3)
plotClust(dat.scaled, dat.scaled$hclust3)
ggsave(file="out/cluster_hclust3.png", width=iwidth, height=iheight)
dat.scaled$hclust8 = cutree(clusters, 8)
plotClust(dat.scaled, dat.scaled$hclust8)
ggsave(file="out/cluster_hclust8.png", width=iwidth, height=iheight)
```