ROI求解器多参数目标函数

Question

我正在尝试为 g40 和 g60 找到最佳解决方案，例如 change=0.01 下面是我的代码，但是我收到错误 Error in .check_function_for_sanity(F, n) ： 无法使用 'n' = 2 个参数评估函数 'F'

我试图寻找解决方案，我发现我应该定义一个包装函数，但它仍然不起作用。 我是 R 优化新手，如果有其他错误，请不要犹豫，教我。

以下是我的代码（数据文件包含两列：p 和 L）

#weights
p40=which(abs(data$p-0.4)==min(abs(data$p-0.4)))                #position of the 40th percentile
w40=data[p40,2]
w60=1-w40

#Non-cumulative L
L_non=data$L[1]
for(i in 2:nrow(data))
L_non[i]=data$L[i]-data$L[i-1]

#X

x1=sum(data$p*L_non)/sum(L_non)
x2=sum(data$p[1:p40]*L_non[1:p40])/sum(L_non[1:p40])
                #Define function for the optimisation               
                
FG=function(g40,g60) {
#Gammas and m
gama=g40*w40+g60*w60
gama40=g40
m=gama40-gama

#theta & delta
theta=-m/(x2-x1)
delta=gama+theta*x1

#L_prime
L_prime=(1+delta-theta*data$p)*L_non
data=cbind(data,L_prime)

#Gini approximated from the new distribution
Lp0.2=data[which(abs(data$p-0.2)==min(abs(data$p-0.2))),3]
Lp0.4=data[which(abs(data$p-0.4)==min(abs(data$p-0.4))),3]
Lp0.6=data[which(abs(data$p-0.6)==min(abs(data$p-0.6))),3]
Lp0.8=data[which(abs(data$p-0.8)==min(abs(data$p-0.8))),3]

Gini_prime=2*(125/288)*(1/5)*(3*(0.2-Lp0.2)+2*(0.4-Lp0.4)+2*(0.6-Lp0.6)+3*(0.8-Lp0.8))

#change
change=(Gini_prime/Gini)-1
}

                # Check that the F_objective function works:
wrapper <- function(x) FG(x, 0.1)

                #OP problem
library(ROI)
library(ROI.plugin.nloptr)

prob<-OP(objective=F_objective(F = wrapper, n = 2),
         constraints=F_constraint(sum(L_prime), dir=c("=",),rhs=c(1)) & L_constraint(change, dir=c("=",),rhs=c(0.01))
         )
solve=ROI_solve(prob,solver="nloptr.cobyla",start=c(0,0))
H=solution(solve)

Answer 1

#weights
p40=which(abs(data$p-0.4)==min(abs(data$p-0.4)))                #position of the 40th percentile
w40=data[p40,2]
w60=1-w40

#Non-cumulative L
L_non=data$L[1]
for(i in 2:nrow(data))
L_non[i]=data$L[i]-data$L[i-1]

#X

x1=sum(data$p*L_non)/sum(L_non)
x2=sum(data$p[1:p40]*L_non[1:p40])/sum(L_non[1:p40])

###################################  Define Optimization Functions  ###################################             
                
    FG=function(g) {      # Objective function
    #Gammas and m
    gama=g[1]*w40+g[2]*w60
    gama40=g[1]
    m=gama40-gama
    return(m)
    }


    FP=function(g) {           #sum of L_prime
    #Gammas and m
    gama=g[1]*w40+g[2]*w60
    gama40=g[1]
    m=gama40-gama

    #theta & delta
    theta=-m/(x2-x1)
    delta=gama+theta*x1

    #L_prime
    L_prime=(1+delta-theta*data$p)*L_non
    return(sum(L_prime))
    }


CH=function(g){       #Gini calculator
#Gammas and m
gama=g[1]*w40+g[2]*w60
gama40=g[1]
m=gama40-gama

#theta & delta
theta=-m/(x2-x1)
delta=gama+theta*x1

#L_prime
L_prime=(1+delta-theta*data$p)*L_non

#Cumulative L
LC_prime=NULL
LC_prime[1]=L_prime[1]
for(i in 2:length(L_prime)) LC_prime[i]=LC_prime[i-1]+L_prime[i]

#Gini approximated from the new distribution

Lp0.2=LC_prime[which(abs(data$p-0.2)==min(abs(data$p-0.2)))]
Lp0.4=LC_prime[which(abs(data$p-0.4)==min(abs(data$p-0.4)))]
Lp0.6=LC_prime[which(abs(data$p-0.6)==min(abs(data$p-0.6)))]
Lp0.8=LC_prime[which(abs(data$p-0.8)==min(abs(data$p-0.8)))]

Gini_prime=2*(125/288)*(1/5)*(3*(0.2-Lp0.2)+2*(0.4-Lp0.4)+2*(0.6-Lp0.6)+3*(0.8-Lp0.8))

#change
# change=(Gini_prime/Gini)-1
return(Gini_prime)
}



###################################  OP problem ################################### 

library(ROI)
library(ROI.plugin.nloptr)

prob<-OP(objective=F_objective(F = FG, n = 2),
         constraints=F_constraint(list(F=CH,F=FP), dir=c("<=","<="),rhs=c(Gini*growth,1)),
         types=c("C","C"),
         bounds = V_bound(ui = seq_len(2), lb = c(-5,-5),ub=c(5,5))
         )
solve=ROI_solve(prob,solver="nloptr.cobyla",start=c(0,0))
G=solution(solve)