多项逻辑/模型选择[Net徽标]

Question

我用多项式logit模型作为概率函数来模拟一组特定代理的选择，当我计算了P1、P2和P3的概率以及如何使用它们来实际建模选择时，我的问题就出现了。我的想法是使用类似于模型库中的中奖例子，但问题是，在这种情况下，概率不是设定的，而是对每一个滴答的变化，即使P1有最低的机会发生，它在下一次迭代中可能有最高的机会。谢谢你的帮助。

ps。多项式逻辑的Eq = {exp(utilityik) \sum exp(utilityjk)}。

to flightchoice-business

let flight-time1 mean [time1] of airline1
let airfare-airline1 mean [airfare] of airline1

set beta1 .8
set beta2 .2

set utility1 beta1 * (airfare-airline1) + beta2 * abs((time - flight-time1) * 10)

let flight-time2 mean [time1] of airline2
let airfare-airline2 mean [airfare] of airline2

set beta1 .8
set beta2 .2

set utility2 beta1 * (airfare-airline2) + beta2 * abs((time - flight-time2) * 10)

let flight-time3 mean [time1] of airline3
let airfare-airline3 mean [airfare] of airline3

set beta1 .8
set beta2 .2

set utility3 beta1 * (airfare-airline3) + beta2 * abs((time - flight-time3) * 10)

let cumulsum exp(utility1) + exp(utility2) + exp(utility3)

let P1 exp(utility1) / cumulsum

let P2 exp(utility2) / cumulsum

let P3 exp(utility3) / cumulsum

let buy random-float P1 + P2 + P3




end

Answer 1

你走在正确的轨道上。您的代码应继续如下：

...
let buy random-float P1 + P2 + P3
ifelse buy < P1 [
  do-stuff-with-P1
] [
  ifelse buy < P1 + P2 [
    do-stuff-with-P2
  ] [
    do-stuff-with-P3
  ]
]

您可以使用一个包含列表的函数来概括这一点，在我看来，它更好一些：

to-report weighted-random [ weights ]
  let pick random-float sum weights
  let total 0
  let i 0
  foreach weights [
    set total total + ?
    if pick < total [ report i ]
    set i i + 1
  ]
end

您可以如下所示：

let buy weighted-random (list P1 P2 P3)
if buy = 0 [ do-stuff-with-P1 ]
if buy = 1 [ do-stuff-with-P2 ]
if buy = 2 [ do-stuff-with-P3 ]