"Baseline" performance measure: Creates an objective function that performs
normal parameter optimization by evaluating filters with additional parameters:
mosmafs.nselect
(how many features to select),
mosmafs.iselect
(vector integer parameter that selects explicit features
that are not necessary the best according to filter values)
and mosmafs.select.weights
(numeric parameter vector that does
weighting between filter values to use.
makeBaselineObjective(
learner,
task,
filters,
ps,
resampling,
measure = NULL,
num.explicit.featsel = 0,
holdout.data = NULL,
worst.measure = NULL,
cpo = NULLCPO,
numfeats = getTaskNFeats(task)
)
function
that can be used for mlrMBO; irace possibly needs some
adjustments.
[Learner]
the base learner to use.
[Task]
the task to optimize.
[character]
filter values to evaluate and use.
[ParamSet]
the ParamSet of the learner to evaluate. Should
not include selector.selection
etc., only parameters of the actual
learner.
[ResampleDesc | ResampleInstance]
the resampling strategy to use.
[Measure]
the measure to evaluate.
If measure needs to be maximized, the measure is multiplied by -1,
to make it a minimization task.
[integer(1)]
additional number of parameters
to add for explicit feature selection.
[Task | NULL]
the holdout data to consider.
[numeric(1)]
worst value to impute for failed evals.
[CPO]
CPO pipeline to apply before feature selection.
[integer(1)]
number of features to consider. Is extracted
from the task
but should be given if cpo
changes the number of features.