Random-Forest/rf_data_v3.R at main · DaconPrediction/Random-Forest · GitHub

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
library(lubridate) #날짜 연산 package
setwd("D:/Rstudy/produce_data")
tr <- read.csv("public_data/train.csv")
tr <- tr[,-2]
tr$date <- as.Date(tr$date)

#nmae측정 함수
nm <- function(prediction, answer){
  nmae <- abs((prediction-answer)/answer)
  out <- c() #Inf인 값들의 index
  for(i in 1:38){
    if(nmae[i]==Inf) {
      out <- c(out, i)
    }
  }
  nmae <- nmae[-c(out)]
  score <- rowMeans(nmae)
  return(score)
}

######train데이터 준비######
#n=1~21-> 품목 개수 (1주뒤 예측) 모델 return
#train할 날짜
date <- c(seq(as.Date('2017/09/28','%Y/%m/%d'),
              as.Date('2017/12/01','%Y/%m/%d'),1),
          seq(as.Date('2018/09/28','%Y/%m/%d'),
              as.Date('2018/12/01','%Y/%m/%d'),1),
          seq(as.Date('2019/09/28','%Y/%m/%d'),
              as.Date('2019/12/01','%Y/%m/%d'),1))

tr_date <- matrix(rep(0,len=6045),nrow=31,ncol=195) #날짜data포함
tr_date <- as.data.frame(tr_date)
colnames(tr_date) <- date
#train할 날짜데이터를 가지는 DF->tr_date
wk <- 2 #wk주 뒤 결과
for(i in 1:195){
  a <- c()
  day <- c()
  day <- date[i]
  a <- date[i]+7*wk #예측할 label값(wk주뒤)
  a <- c(a,day-years(1)) #1년전 데이터
  a <- c(a, day-seq(56,84,1))
  tr_date[,i] <- a
}
rownames(tr_date) <- c('label','1year',56:84)
tr_date <- t(tr_date)
tr_date <- as.data.frame(tr_date)
tr_price <- matrix(rep(0,len=6045),nrow=195,ncol=31) #가격데이터를 담은 DF
tr_price <- as.data.frame(tr_price)
#train할 가격데이터를 가지는 DF->tr_price

n <- 2 #농산물 번호(순서대로)

for(i in 1:195){
  for(j in 1:31){
    tr_price[i,j] <- tr[tr_date[i,j]==tr$date,1+2*n]
  }
}
subset <- subset(tr,tr[,1+2*n]!=0)
laplace <- mean(subset[,1+2*n]) #원하는 농작물행의 평균값
for(i in 1:195){
  for(j in 1:31){
    #0인 값들을 각 농작물의 평균으로 대체
    if(tr_price[i,j]==0){
      tr_price[i,j] <- laplace
    }
  }
}
colnames(tr_price) <-c('label','1year',56:84)
rownames(tr_price) <- date

subset1 <- subset(tr_price$label,tr_price$label!=0)
laplace1 <- mean(subset1)
for(i in 1:length(tr_price$label)){
  if(tr_price$label[i]==0){
    tr_price$label[i] <- laplace1
  }
}

#랜덤 포레스트
library(randomForest)
set.seed(300)
attach(tr_price)
model <- randomForest(label~.,  data = tr_price, mtry = floor(sqrt(30)), ntree = 500) #랜덤 포레스트 모델 생성
summary(model)

m <- model

pred <- predict(m,tr_price[,2:31])
print(cor(pred,tr_price$label))


######test데이터 준비######
library(psych)#산포도와 상관관계, 히스토그램 등을 모두 보여주는 함수
library(MASS)
date2 <- seq(as.Date('2020/09/29','%Y/%m/%d'), #test할 데이터의 날짜
             as.Date('2020/11/05','%Y/%m/%d'),1)

ts_date <- matrix(rep(0,len=1178),nrow=31,ncol=38) #날짜data포함
ts_date <- as.data.frame(ts_date)
colnames(ts_date) <- date2
#train할 날짜데이터를 가지는 DF
for(i in 1:38){
  b <- c()
  day <- c()
  day <- date2[i]
  b <- date2[i]+7*wk #예측할 label값(wk주뒤)
  b <- c(b,day-years(1)) #1년 전 데이터
  b <- c(b, day-seq(56,84,1))
  ts_date[,i] <- b
}
rownames(ts_date) <- c('label','1year',56:84)
ts_date <- t(ts_date)

ts_price <- matrix(rep(0,len=1178),nrow=38,ncol=31) #가격데이터를 담은 DF
ts_price <- as.data.frame(ts_price)
for(i in 1:38){
  for(j in 2:31){
    ts_price[i,j] <- tr[ts_date[i,j]==tr$date,1+2*n]
  }
}
subset <- subset(tr,tr[,1+2*n]!=0)
laplace <- mean(subset[,1+2*n])
for(i in 1:38){
  for(j in 1:31){
    #값이 0인 부분을 평균값으로 대체
    if(ts_price[i,j]==0){
      ts_price[i,j] <- laplace
    }
  }
}

colnames(ts_price) <-c('label','1year',56:84)
rownames(ts_price) <- date2

ml <- model
pred <- predict(ml,ts_price[2:31]) #label 체크하기

mx <- matrix(rep(0,len=38),ncol=38,nrow=1)
mx <- as.data.frame(mx)
rownames(mx) <- c("무")
colnames(mx) <- date2+7*wk
mx[1,] <- pred
View(mx)


real_date <- date2+7
real_price <- subset(tr, date %in% real_date) #기간내의 음식들의 실제 가격
View(real_price)


NM <- c() #nmae값 담기
NM <- c(NM,nm(mx[n,],real_price[,2*n+1]))

mean(NM)