Container for results of hyperparameter tuning. Contains the obtained point in search space, its performance values and the optimization path which lead there.
Object members:
- learner (Learner)
Learner that was optimized.
- control (TuneControl)
Control object from tuning.
- x (list)
Named list of hyperparameter values identified as optimal. Note that when you have trafos on some of your params,
x
will always be on the TRANSFORMED scale so you directly use it.- y (numeric)
Performance values for optimal
x
.- threshold (numeric)
Vector of finally found and used thresholds if
tune.threshold
was enabled in TuneControl, otherwise not present and henceNULL
.- opt.path (ParamHelpers::OptPath)
Optimization path which lead to
x
. Note that when you have trafos on some of your params, the opt.path always contains the UNTRANSFORMED values on the original scale. You can simply calltrafoOptPath(opt.path)
to transform them, or,as.data.frame{trafoOptPath(opt.path)}
. If mlr optionon.error.dump
isTRUE
,OptPath
will have a.dump
object in itsextra
column which contains error dump traces from failed optimization evaluations. It can be accessed bygetOptPathEl(opt.path)$extra$.dump
.