optsampleentropy/simulation.R at main · SunHyun-Park/optsampleentropy · 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
158
159
160
161
162
163
164
# Library -----------------------------------------------------------------
rm(list=ls())

source("methods.R")

#install.packages('data.table')
#install.packages('caret')
#install.packages('nloptr')
#install.packages('snowfall')

library(data.table)
library(caret)
library(nloptr)
library(snowfall)

# Parameter ---------------------------------------------------------------

args <- as.numeric(commandArgs(trailingOnly = TRUE))
nsim <- 1000
m <- args[1]
C <- args[2]

optim.tol <- 1e-7
optim.iter <- 1e5

# Load data ---------------------------------------------------------------

data <- fread('healthcare-dataset-stroke-data.csv')
nrow(data)
data <- data[,-c(1)]
data <- data[order(data$stroke)]
data <- data[bmi != "N/A"]
data$bmi <- as.numeric(data$bmi)
nrow(data)
data <- data[gender!="Other"]
nrow(data)
data <- data[smoking_status!="Unknown"]
#Note: "Unknown" in smoking_status means that the information is unavailable for this patient
nrow(data)
data$gender <- as.factor(data$gender)
data$ever_married <- as.factor(data$ever_married)
data$work_type <- as.factor(data$work_type)
data$Residence_type <- as.factor(data$Residence_type)
data$smoking_status <- as.factor(data$smoking_status)

# y, s, x
y <- data$stroke
s <- rep(0,length(data$stroke))

dummy <- dummyVars(stroke~., data=data, fullRank=T)
x <- cbind(1,predict(dummy, data))
mean(y)

# Split pilot -----------------------------------------------

n <- length(y)
set.seed(1)
pilot_idx <- sample(1:n, m)
pilot <- list()
pilot$x <- x[pilot_idx,]
pilot$y <- y[pilot_idx]
pilot$beta <- ftn.mse.beta(pilot$y,pilot$x,rep(1,m),rep(1,m),rep(0,ncol(pilot$x)))$par
pilot$beta2 <- ftn.ce.beta(pilot$y,pilot$x,rep(1,m),rep(1,m),rep(0,ncol(pilot$x)))$par

y <- y[-pilot_idx]
s <- s[-pilot_idx]
x <- x[-pilot_idx,]


n <- length(y)
n0 <- sum(1-s)

# FULL --------------------------------------------------------------------

#model.s <- glm(s~.-1-y,data=data.table(y=y,s=s,x=x),family='binomial')
ps=rep(0, n)

full <- list()
full$beta <- ftn.ce.beta(y,x,rep(1,n),rep(1,n),pilot$beta2)$par
full <- c(full,ftn.summary(y,s,x,ps,full$beta))

# w OPT weights -------------------------------------------------------------

pilot$w <- ftn.piopt.mse(s,x,pilot$beta,C)

# pi OPT weights ------------------------------------------------------------

pilot$pi <- ftn.piopt.ce(s, x, pilot$beta2, C)

# Simulation
ftn.sim <- function(iter) {
  set.seed(iter)
  result <- NULL
  ecount <- 0

  while (is.null(result)) {
    result <- tryCatch({

      dataset <- data.frame(y=y, s=s, x)

      # SRS ------------------------------------------------------------
      w <- rep(C, n0) / n0
      r <- rep(0, n)
      r[sample(1:n0, C)] <- 1

      # MLE under SRS --------------------------------------------------
      srs <- list()
      srs$est <- ftn.mle.beta(y, x, r, pilot$beta2)
      if(srs$est$convergence!=0) stop("srs convergence error")
      srs <- c(srs, ftn.summary(y, s, x, ps, srs$est$par))

      # Minimize MSE under OPT------------------------------------------
      optw <- list()
      optw$w <- pilot$w
      optw$r <- rep(0, n)
      optw$r[which(rbinom(n0, 1, optw$w) == 1)] <- 1
      optw$est <- ftn.mse.beta(y, x, optw$r, optw$w, pilot$beta)
      if(optw$est$convergence!=0) stop("optw convergence error")
      optw <- c(optw, ftn.summary(y, s, x, ps, optw$est$par))

      # Minimize Cross-Entropy Loss under OPT --------------------------
      optpi <- list()
      optpi$pi <- pilot$pi
      optpi$r <- rep(0, n)
      optpi$r[which(rbinom(n0, 1, optpi$pi) == 1)] <- 1
      optpi$est <- ftn.ce.beta(y, x, optpi$r, optpi$pi, pilot$beta2)
      if(optpi$est$convergence!=0) stop("optpi convergence error")
      optpi <- c(optpi, ftn.summary(y, s, x, ps, optpi$est$par))

      # OSCA -----------------------------------------------------------
      osca <- list()
      osca$beta <- ftn.osca(y,s,x,r,rep(0, ncol(x)))
      osca <- c(osca,ftn.summary(y,s,x,ps,osca$beta))

      # Summary --------------------------------------------------------
      list(
        full.p = full$p,  optw.p = optw$p,  optpi.p = optpi$p, osca.p = osca$p, srs.p = srs$p,
        full.mse = full$mse, optw.mse = optw$mse, optpi.mse = optpi$mse, osca.mse = osca$mse, srs.mse = srs$mse,
        # optw.w = optw$w, optpi.pi = optpi$pi, srs.pi = rep(C / n0, n0),
        ecount = ecount
      )
    }, error = function(e) {
      message(paste("An error occurred - Retrying. Error message:", conditionMessage(e)))
      ecount <<- ecount + 1
      return(NULL)
    })
  }

  return(result)
}

# Parallel computing ------------------------------------------------------

sfInit(parallel=TRUE, cpus=32, type="SOCK")
sfExportAll()
result <- sfClusterApplyLB(1:nsim, ftn.sim)
sfStop()

# Output ------------------------------------------------------------------

save.image(file=paste0('result-',m,'-',C,'.RData'))