forked from RubensZimbres/Repo-2016
-
Notifications
You must be signed in to change notification settings - Fork 0
/
R - Ridge Regularization
174 lines (145 loc) · 4.67 KB
/
R - Ridge Regularization
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
closeAllConnections()
rm(list=ls())
library(parallel)
setwd("/Volumes/16 DOS/R_nbs")
sampleML<-read.csv("GameR.csv",sep=",",header=TRUE)
head(sampleML)
library(lattice)
attach(mtcars)
par(mfrow=c(1,1))
plot(sort(sampleML[1:2879,3]))
4053-sum(is.na(sampleML$y)==TRUE)
plot(y~x2,data=sampleML)
abline(lm(y~x2,data=sampleML),col="red")
boxplot(y~x2,data=sampleML[100:130,])
hist(sampleML$y,data=sampleML$y)
#Normal
qqnorm(sampleML$y)
qqline(sampleML$y,col="blue",lty=1)
shapiro.test(sampleML$y)
#missing
require(mice)
md.pattern(sampleML)
require(VIM)
imputation<-mice(sampleML,m=5,maxit=10,method='pmm',seed=500)
completeDataML<-complete(imputation,2)
completeDataML
#outliers
require(DMwR)
outliers.score<-lofactor(completeDataML,k=50)
outliers <- order(outliers.score, decreasing=T)[1:30]
outliers
delete=list(outliers)
t<-c(outliers)
filteredGame<-completeDataML[-t,]
dim(completeDataML)
dim(filteredGame)
#Linear Regression
x_train<-filteredGame[1:2700,1:2]
y_train<-filteredGame[1:2700,3]
x_test<-filteredGame[1:2700,1:2]
x<-cbind(x_train,y_train)
y_train
cor(sampleML[1:2879,])
par(mfrow=c(1,1))
plot(x_train[[1]],y_train)
plot(x_train[[2]],y_train)
linear<-lm(y_train~x_train[[1]]+x_train[[2]]+I(x_train[[2]]^2),data=filteredGame)
summary(linear)
yy0<-c()
for (i in seq(.6,2.5,0.001)){
yy0<-append(yy0,1-mean(abs(y_train-lm(y_train~x_train[[1]]+x_train[[2]]+I(x_train[[2]]^i),data=filteredGame)$fitted.values)),after=length(yy0))
}
max(yy0)
which(yy0==max(yy0))
coef0<-seq(.6,2.5,0.001)[[which(yy0==max(yy0))]]
coef0
linear<-lm(y_train~x_train[[1]]+x_train[[2]]+I(x_train[[2]]^coef0),data=filteredGame)
summary(linear)
hist(linear$residuals)
1-mean(abs(linear$residuals))
hist(log(x_train[[1]]))
hist(log(x_train[[2]]))
hist(log(y_train))
# NON LINEAR REGRESSION WITH NORMAL DATA
yy0<-c()
for (i in seq(.6,2.5,0.001)){
yy0<-append(yy0,1-mean(abs(log(y_train)-lm(log(y_train)~log(x_train[[1]])+log(x_train[[2]])+I(log(x_train[[2]])^i),data=filteredGame)$fitted.values)),after=length(yy0))
}
max(yy0)
which(yy0==max(yy0))
coef0<-seq(.6,2.5,0.001)[[which(yy0==max(yy0))]]
coef0
# NON LINEAR ALL ADJUSTED
# 0.71 para 0.76 R2
linear2<-lm(log(y_train)~log(x_train[[1]])+log(x_train[[2]])+I(log(x_train[[2]])^coef0),data=filteredGame)
summary(linear2)
hist(linear2$residuals)
names(linear2)
accuracy<-1-mean(abs(log(y_train)-linear2$fitted.values))
accuracy
plot(log(y_train),linear2$fitted.values)
barchart(linear2$coefficients)
hist(linear2$residuals)
### CERTO
par(mfrow=c(1,1))
ccc<-cbind(log(sampleML[1:2700,2]),log(sampleML[1:2700,3]),linear2$fitted.values)
head(ccc)
ddd<-setorder(data.frame(ccc))
as.numeric(ddd[,2])
plot(as.numeric(ddd[,1]),as.numeric(ddd[,2]),col="black",pch=20,xlab="Cases",ylab="Y")
points(as.numeric(ddd[,1]),as.numeric(ddd[,3]),col="red",pch=20)
sort(ccc,2,decreasing=FALSE)
plot(sort(log(sampleML[1:2700,3])),col="black",pch=20,xlab="Cases",ylab="Y")
points(sort(linear2$fitted.values),col="red",pch=20)
# REGULARIZATION
lambda=0.001
linear2$coefficients1<-linear2$coefficients[1]-
lambda*sum((log(y_train)-linear2$fitted.values)^2)
linear2$coefficients2<-linear2$coefficients[2]-
lambda*sum((log(y_train)-linear2$fitted.values)^2)
linear2$coefficients3<-linear2$coefficients[3]-
lambda*sum((log(y_train)-linear2$fitted.values)^2)
linear2$coefficients4<-linear2$coefficients[4]-
lambda*sum((log(y_train)-linear2$fitted.values)^2)
bb<-linear2$coefficients[1]+linear2$coefficients2*log(x_train[[1]])+
linear2$coefficients3*log(x_train[[2]])+linear2$coefficients4*
log(x_train[[2]])^coef0
bb
par(mfrow=c(1,1))
plot(sort(log(sampleML[1:2700,3]),decreasing=TRUE),col="black",pch=20,xlab="weight",ylab="Error")
points(1+sort(bb,decreasing=TRUE),col="green",pch=20)
library("ggplot2")
p1<-ggplot(sampleML[1:2700,], aes(x=sort(log(sampleML[1:2700,3])),y=c(1:2700)),color="black")+geom_line(size=1.5)
p1
p1+geom_line(data=sampleML[1:2700,], aes(x=1+sort(bb),y=c(1:2700)),size=2,color="green")
p1+geom_smooth(color="green")+
xlab("Error") +ylab("weight") +
ggtitle(expression(atop("Hill Climbing", atop(italic("Regularization"), ""))))
x3<-log(sampleML[1:2700,2])
y3<-log(sampleML[1:2700,3])
c2<-ggplot(sampleML[1:2700,], aes(x3,y3))+geom_point()
c2
c2 + stat_smooth(method = "lm",color="red") + geom_point()
# ANOVA
#ANOVA
fit <- aov(linear2, data=sampleML)
summary(fit)
# PREDICTION
x_train<-filteredGame[1:500,1:2]
x_train[[1]][x_train[[1]]==0]<-0.01
y_train<-filteredGame[1:500,3]
x<-cbind(x_train,y_train)
y_train
pred<-predict(linear2,newdata=x)
# COMO TRANSFORMAR LOG EM NUMERO CERTO = ^10
prev<-as.numeric(10^pred)
pp<-prev+.26
y_train-pp
acc<-1-mean(abs(y_train-(prev+.26)))
acc
plot(y_train,pp)
y_train
write.csv(prev, "PredictedNonLinearOK.csv")
print(c("ACCURACY",acc))
hist(y_train-pp)