How to handle "ineligible" learner parameters?

[MariaErdmann]

RWeka's J48, for example, has a parameter "output-debug-info". Hyphens are not allowed in R. This causes problems when using generateDesignOfDefaults:

ps = getParamSet("classif.J48")

# Modify so that ps has finite box constraints
ps.def = filterParams(ps, ids = names(getDefaults(ps)))
finite.constraints = lapply(ps.def$pars, function(p) {
  if (p$type %in% c("integer", "numeric")) {
    p$upper = ifelse(p$upper == Inf, p$default, p$upper)
    p$lower = ifelse(p$lower == -Inf, p$default, p$lower)
  }
  return(p)
})

ps.finite.constraints = do.call(makeParamSet, finite.constraints)
generateDesignOfDefaults(ps.finite.constraints)

[mb706]

This is a bug in ParamHelpers, you should report it there...

By default, invalid parts of a variable name should be replaced by dots, so in the design data frame, the column should be named output.debug.info.

A workaround seems to be to remove the parameter from the param set to generate the default dataframe, and then to add the parameter manually:

ps.finite.constraints2 <- ps.finite.constraints
ps.finite.constraints2$pars[['output-debug-info']] <- NULL
desi = generateDesignOfDefaults(ps.finite.constraints2)
desi$output.debug.info=getDefaults(ps.finite.constraints)[['output-debug-info']]
ctrl4 <- makeTuneControlDesign(design=desi)
tuneParams("classif.J48", pid.task, resampling=cv3, control=ctrl4, par.set=ps.finite.constraints, show.info=TRUE)

[stale[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.