选择MLR遗传算法的初始种群

Question

我想使用 mlr 包中的 makeFeatSelControlGA() 函数选择用于变量选择的遗传算法的初始种群。可行吗？

编辑

由于未实现，我为此对 selectFeaturesGA 函数进行了脏编辑。这是代码，如果它对任何人有帮助的话。

library(R.utils)
reassignInPackage("selectFeaturesGA","mlr",function (learner, task, resampling, measures, bit.names, bits.to.features, 
                                                     control, opt.path, show.info) 
{
  states=get("states",envir=.GlobalEnv)
  mu = control$extra.args$mu
  lambda = control$extra.args$lambda
  yname = opt.path$y.names[1]
  minimize = opt.path$minimize[1]

  if (!length(states))
  {
    for (i in seq_len(mu)) {
      while (TRUE) {
        states[[i]] = rbinom(length(bit.names), 1, 0.5)
        if (is.na(control$max.features) || sum(states[[i]] <= 
                                               control$max.features)) 
          break
      }
    }
  }
  
  evalOptimizationStatesFeatSel(learner, task, resampling, 
                                measures, bits.to.features, control, opt.path, show.info, 
                                states, 0L, NA_integer_)
  pop.inds = seq_len(mu)
  for (i in seq_len(control$maxit)) {
    pop.df = as.data.frame(opt.path)[pop.inds, , drop = FALSE]
    pop.featmat = as.matrix(pop.df[, bit.names, drop = FALSE])
    mode(pop.featmat) = "integer"
    pop.y = pop.df[, yname]
    kids.list = replicate(lambda, generateKid(pop.featmat, 
                                              control), simplify = FALSE)
    kids.evals = evalOptimizationStatesFeatSel(learner, task, 
                                               resampling, measures, bits.to.features, control, 
                                               opt.path, show.info, states = kids.list, i, as.integer(NA))
    kids.y = extractSubList(kids.evals, c("y", yname))
    oplen = getOptPathLength(opt.path)
    kids.inds = seq(oplen - lambda + 1, oplen)
    if (control$extra.args$comma) {
      setOptPathElEOL(opt.path, pop.inds, i - 1)
      pool.inds = kids.inds
      pool.y = kids.y
    }
    else {
      pool.inds = c(pop.inds, kids.inds)
      pool.y = c(pop.y, kids.y)
    }
    pop.inds = pool.inds[order(pool.y, decreasing = !minimize)[seq_len(mu)]]
    setOptPathElEOL(opt.path, setdiff(pool.inds, pop.inds), 
                    i)
  }
  makeFeatSelResultFromOptPath(learner, measures, resampling, 
                               control, opt.path)
})

并且在全局环境中，需要添加一个名为states的列表，长度等于mu

Answer 1

mlr 目前不支持此功能，因此您必须修改源代码才能实现。

选择MLR遗传算法的初始种群

Choosing initial population of mlr genetic algorithm

r

genetic-algorithm

mlr

编辑