-
Notifications
You must be signed in to change notification settings - Fork 0
/
LSED_Lab_4__Zadanie_3.R
150 lines (111 loc) · 5.23 KB
/
LSED_Lab_4__Zadanie_3.R
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
### LSED - Zadanie 3 - LAB nr 4
# Autor: Szymon Baczyqski
rm(list=ls())
library(MASS)
library(class)
draw.data.gauss <- function(S1, S2, m1, m2, n1, n2) {
X1 <- mvrnorm(n1, m1, S1)
X2 <- mvrnorm(n2, m2, S2)
X1 <- data.frame(X1); colnames(X1) <- c("x", "y")
X2 <- data.frame(X2); colnames(X2) <- c("x", "y")
X1$class <- 1; X2$class <- 2
data <- rbind(X1, X2); data$class <- factor(data$class)
return(data)
}
CM.large <- function(org.class, pred.class) {
CM <- table(org.class, pred.class)
# Skuteczno f klasyfikatora
ACC <- sum(diag(CM)) / sum(CM)
# Warto ci true positive i true negative
# zak3adamy, ?e klasa "2" jest "pozytywna"
TP <- CM[2,2]
TN <- CM[1,1]
sums <- apply(CM, 1, sum)
TPR <- TP / sums[2]
FPR <- 1 - TN / sums[1]
return(c(ACC = round(ACC,4), TP = TP, TN = TN, row.names = NULL))
}
### --- Czj f 1 --- wykre lif skuteczno f klasyfikatora knn w funkcji liczby najbli?szych s9siadsw od knn=1 do knn=21. Wykonac to samo dla warto ci TP i TN. --- ###
# Parametry danych z rozkladu Gaussa
S1 <- matrix(c(4, 2, 2, 4), 2, 2)
S2 <- matrix(c(4, 2, 2, 2), 2, 2)
m1 <- c(-1, -1)
m2 <- c(2, 2)
n1 <- 30
n2 <- 20
# Generowanie danych
data <- draw.data.gauss(S1, S2, m1, m2, n1, n2)
knn.val <- CM.large(data$class, class.knn <- knn(data[,1:2], data[,1:2], data$class, 1))
k_max <- 21
for(i in 2:k_max){
knn.val <- rbind(knn.val, CM.large(data$class, class.knn <- knn(data[,1:2], data[,1:2], data$class, i)))
}
rownames(knn.val) <- c(1:k_max)
plot(x=1:k_max, y=knn.val[,1], t="b", col="blue", xlab = "k_nn", ylab = "ACC", xlim=c(1,k_max), pch=19)
title("ACC(k_nn) dla k_nn = 1:21")
axis(side=1, at=c(1:k_max))
plot(x=1:k_max, y=knn.val[,2], t="b", ylim=c((min(knn.val[,2])),(max(knn.val[,3]))),xlab = "k_nn", ylab = "TP & TN", xlim=c(1,k_max), pch=19, col="blue")
title("TP i TN dla k_nn = 1:21")
axis(side=1, at=c(1:k_max))
#axis(side=2, tick = 1)
points((1:k_max), knn.val[,3], t="b", pch=19, col="red")
legend("topright", legend=c("TN (klasa 1)","TP (klasa 2)"), col=c("red","blue"), lty=1:2, cex=1)
### --- Czesc 2 --- Nastjpnie wylosowaf dodatkowo 10 punktsw z klasy 1 oraz 5 punktsw z klasy 2, potraktowaf jako zbisr tetsowy i powtsrzyf wykresy. --- ###
### --- --- W razie mozliwosci doknac usrednienia po 10 losowaniach. --- ###
n1t <- 10
n2t <- 5
test <- draw.data.gauss(S1, S2, m1, m2, n1t, n2t)
test.val <- CM.large(test$class, classtest.knn <- knn(data[,1:2], test[,1:2], data$class, 1))
for(i in 2:k_max){
test.val <- rbind(test.val, CM.large(test$class, classtest.knn <- knn(data[,1:2], test[,1:2], data$class, i)))
}
rownames(test.val) <- c(1:k_max)
plot(x=1:k_max, y=test.val[,1], t="b", col="blue", xlab = "k_nn", ylab = "TEST ACC", xlim=c(1,k_max), pch=19)
title("TEST ACC(k_nn) dla k_nn = 1:21")
axis(side=1, at=c(1:k_max))
plot(x=1:k_max, y=test.val[,2], t="b", ylim=c((min(test.val[,2])),(max(test.val[,3]))),xlab = "k_nn", ylab = "TEST: TP & TN", xlim=c(1,k_max), pch=19, col="blue")
title("TEST: TP i TN dla k_nn = 1:21")
axis(side=1, at=c(1:k_max))
#axis(side=2, tick = 1)
points((1:k_max), test.val[,3], t="b", pch=19, col="red")
legend("right", legend=c("TN (klasa 1)","TP (klasa 2)"), col=c("red","blue"), lty=1:2, cex=1)
### --- Czesc 2.1 --- W razie moziwosci doknac usrednienia po 10 losowaniach. --- ###
# Usrednianie po 10 losowaniach
los <- 10
ACC_mean = 0; TP_mean = 0; TN_mean = 0;
for(x in 1:los){
test.val <- CM.large(test$class, classtest.knn <- knn(data[,1:2], test[,1:2], data$class, 1))
for(i in 2:k_max){
test.val <- rbind(test.val, CM.large(test$class, classtest.knn <- knn(data[,1:2], test[,1:2], data$class, i)))
}
rownames(test.val) <- c(1:k_max)
ACC_mean <- cbind(ACC_mean,test.val[,1])
TP_mean <- cbind(TP_mean,test.val[,2])
TN_mean <- cbind(TN_mean,test.val[,3])
}
ACC_mean <- ACC_mean[,-1]; TP_mean <- TP_mean[,-1]; TN_mean <- TN_mean[,-1]
para.val <- data.frame(rowMeans(ACC_mean), round(rowMeans(TP_mean)), round(rowMeans(TN_mean)), apply(ACC_mean, 1,sd), apply(TP_mean, 1,sd), apply(TN_mean, 1,sd))
# --- Plot ACC od k_nn
plot(x=1:k_max, y=para.val[,1], t="b", col="blue", xlab = "k_nn", ylab = "TEST ACC", xlim=c(1,k_max), pch=19)
title("U REDNIONY TEST ACC(k_nn) dla k_nn = 1:21")
axis(side=1, at=c(1:k_max))
# --- Plot TP i TN od k_nn
plot(x=1:k_max, y=para.val[,2], t="b", ylim=c((min(para.val[,2])),(max(para.val[,3]))),xlab = "k_nn", ylab = "U REDNIONY TEST: TP & TN", xlim=c(1,k_max), pch=19, col="blue")
title("U REDNIONY TEST: TP i TN dla k_nn = 1:21")
axis(side=1, at=c(1:k_max))
points((1:k_max), para.val[,3], t="b", pch=19, col="red")
legend("right", legend=c("TN (klasa 1)","TP (klasa 2)"), col=c("red","blue"), lty=1:2, cex=1)
library(Hmisc)
K_nn = 1:k_max
#ACC = para.val[,1]
#TP = para.val[,2]
#TN = para.val[,3]
#SD(ACC_rows) = para.val[,4] - dla ka?dego K_nn
#SD(TP_rows) = para.val[,5]
#SD(TN_rows) = para.val[,6]
errbar(K_nn, para.val[,1], (para.val[,1] + para.val[,4]), (para.val[,1] - para.val[,4]), ylab = "ACC Mean")
title("ACC Mean")
errbar(K_nn, para.val[,2], (para.val[,2] + para.val[,5]), (para.val[,2] - para.val[,5]), ylab = "TP Mean (Klasa 2)")
title("TP Mean (Klasa 2 - po u rednianiu)")
errbar(K_nn, para.val[,3], (para.val[,3] + para.val[,6]), (para.val[,3] - para.val[,6]), ylab = "TN Mean (Klasa 1)")
title("TN Mean (Klasa 1 - po u rednianiu)")