Question

我有以下数据：

set.seed(1)
data <- data.frame(matrix(rexp(45), 5))
data <- cbind(data, c(1,2,2,3,1))
colnames(data) <- c("model1_a","model1_b","model1_c","model2_a","model2_b","model2_c","model3_a","model3_b","model3_c", "true_max")

那是：

   model1_a  model1_b  model1_c  model2_a  model2_b   model2_c   model3_a  model3_b  model3_c true_max
1 0.7551818 2.8949685 1.3907351 1.0352439 2.3645153 0.05943916 1.43528534 1.0227259 1.0798811        1
2 1.1816428 1.2295621 0.7620299 1.8760352 0.6418926 0.57871246 0.03726853 0.3017409 1.0282469        2
3 0.1457067 0.5396828 1.2376036 0.6547466 0.2941204 3.95893285 0.32401015 0.7252143 1.2922616        2
4 0.1397953 0.9565675 4.4239342 0.3369335 0.5658655 1.17331211 1.32046793 0.7515427 1.2531054        3
5 0.4360686 0.1470460 1.0545432 0.5884797 0.1060726 0.99681296 0.20351035 0.2350275 0.5546414        1

使用这些数据，我需要找到最佳权重来组合每个模型的结果。没有权重，这些总和是：

  cbind(
    "a_total" = data$model1_a+data$model2_a+data$model3_a,
    "b_total" = data$model1_b+data$model2_b+data$model3_b,
    "c_total" = data$model1_c+data$model2_c+data$model3_c
  )

      a_total   b_total  c_total
[1,] 3.225711 6.2822097 2.530055
[2,] 3.094946 2.1731956 2.368989
[3,] 1.124464 1.5590175 6.488798
[4,] 1.797197 2.2739757 6.850352
[5,] 1.228059 0.4881461 2.605998

使用which.max()会产生2 1 3 3 3的估计最大值。如何使用optim()为每列（或单个值）设置权重？数据包含真实的最大值

Answer 1

你可以试试这个：

# weights
# model1_a  model1_b  model1_c  model2_a  model2_b   model2_c   model3_a  model3_b  model3_c
# W1        W2        W3        W4        W5         W6         1-W1-W4   1-W2-W5   1-W3-W6    

# max(data_model_i%*%W_model_i), i=1,2,3
model.opt <- function (W,data) {
  max(sum(data[,c(1,4,7)]*c(W[1],W[4],1-W[1]-W[4])),
      sum(data[,c(2,5,8)]*c(W[2],W[5],1-W[2]-W[5])),
      sum(data[,c(3,6,9)]*c(W[3],W[6],1-W[3]-W[6])))
}

for (i in 1:nrow(data)) {
  W <- optim(par=rep(0.5,6), fn = model.opt, 
          method = "L-BFGS-B",
          lower = rep(0,6), # W_i >= 0 for all i
          upper = rep(1,6), # W_i <= 1 for all i
          control = list(fnscale = -1), 
          data=as.matrix(data[i,]))$par # optimum W
  print(which.max(c(sum(data[i,c(1,4,7)]*c(W[1],W[4],1-W[1]-W[4])),
      sum(data[i,c(2,5,8)]*c(W[2],W[5],1-W[2]-W[5])),
      sum(data[i,c(3,6,9)]*c(W[3],W[6],1-W[3]-W[6])))))
}

[1] 2
[1] 1
[1] 3
[1] 3
[1] 3

R：使用optim（）设置变量权重

1 个答案: