cleanup after unloading

[sebffischer]

After unloading mlr3proba I think the dictionary items that were added by proba should be removed again.
I made a similar PR here already: mlr-org/mlr3learners#258

[RaphaelS1]

Could you copy that PR here?

[jemus42]

This would require to undo everything done in .onLoad here

mlr3proba/R/zzz.R

Lines 41 to 173 in 930d8b0

	register_mlr3 = function() {
	# reflections
	x = utils::getFromNamespace("mlr_reflections", ns = "mlr3")
	# task
	x$task_types = x$task_types[!c("surv", "dens")]
	x$task_types = setkeyv(rbind(x$task_types, rowwise_table(
	~type, ~package, ~task, ~learner, ~prediction, ~prediction_data, ~measure,
	"surv", "mlr3proba", "TaskSurv", "LearnerSurv", "PredictionSurv", "PredictionDataSurv", "MeasureSurv",
	"dens", "mlr3proba", "TaskDens", "LearnerDens", "PredictionDens", "PredictionDataDens", "MeasureDens"
	)), "type")
	x$task_col_roles$surv = x$task_col_roles$regr
	x$task_col_roles$dens = c("feature", "target", "label", "order", "group", "weight", "stratum")
	x$task_properties$surv = x$task_properties$regr
	x$task_properties$dens = x$task_properties$regr
	# learner
	x$learner_properties$surv = x$learner_properties$regr
	x$learner_properties$dens = x$learner_properties$regr
	x$learner_predict_types$surv = list(
	crank = c("crank", "lp", "distr", "response"),
	distr = c("crank", "lp", "distr", "response"),
	lp = c("crank", "lp", "distr", "response"),
	response = c("crank", "lp", "distr", "response"))
	x$learner_predict_types$dens = list(
	pdf = c("pdf", "cdf", "distr"),
	cdf = c("pdf", "cdf", "distr"),
	distr = c("pdf", "cdf", "distr"))
	# measure
	x$measure_properties$surv = x$measure_properties$regr
	x$measure_properties$dens = x$measure_properties$regr
	x$default_measures$surv = "surv.cindex"
	x$default_measures$dens = "dens.logloss"
	# tasks
	x = utils::getFromNamespace("mlr_tasks", ns = "mlr3")
	x$add("precip", load_precip)
	x$add("faithful", load_faithful)
	x$add("rats", load_rats)
	x$add("lung", load_lung)
	x$add("actg", load_actg)
	x$add("gbcs", load_gbcs)
	x$add("grace", load_grace)
	x$add("whas", load_whas)
	x$add("unemployment", load_unemployment)
	# generators
	x = utils::getFromNamespace("mlr_task_generators", ns = "mlr3")
	x$add("simdens", TaskGeneratorSimdens)
	x$add("simsurv", TaskGeneratorSimsurv)
	# learners
	x = utils::getFromNamespace("mlr_learners", ns = "mlr3")
	x$add("dens.hist", LearnerDensHistogram)
	x$add("dens.kde", LearnerDensKDE)
	x$add("surv.coxph", LearnerSurvCoxPH)
	x$add("surv.kaplan", LearnerSurvKaplan)
	x$add("surv.rpart", LearnerSurvRpart)
	# measures
	x = utils::getFromNamespace("mlr_measures", ns = "mlr3")
	x$add("dens.logloss", MeasureDensLogloss)
	x$add("regr.logloss", MeasureRegrLogloss)
	x$add("surv.graf", MeasureSurvGraf)
	x$add("surv.brier", MeasureSurvGraf)
	x$add("surv.schmid", MeasureSurvSchmid)
	x$add("surv.logloss", MeasureSurvLogloss)
	x$add("surv.rcll", MeasureSurvRCLL)
	x$add("surv.intlogloss", MeasureSurvIntLogloss)
	x$add("surv.cindex", MeasureSurvCindex)
	x$add("surv.dcalib", MeasureSurvDCalibration)
	x$add("surv.calib_beta", MeasureSurvCalibrationBeta)
	x$add("surv.calib_alpha", MeasureSurvCalibrationAlpha)
	x$add("surv.nagelk_r2", MeasureSurvNagelkR2)
	x$add("surv.oquigley_r2", MeasureSurvOQuigleyR2)
	x$add("surv.xu_r2", MeasureSurvXuR2)
	x$add("surv.chambless_auc", MeasureSurvChamblessAUC)
	x$add("surv.hung_auc", MeasureSurvHungAUC)
	x$add("surv.uno_auc", MeasureSurvUnoAUC)
	x$add("surv.song_auc", MeasureSurvSongAUC)
	x$add("surv.uno_tpr", MeasureSurvUnoTPR)
	x$add("surv.song_tpr", MeasureSurvSongTPR)
	x$add("surv.uno_tnr", MeasureSurvUnoTNR)
	x$add("surv.song_tnr", MeasureSurvSongTNR)
	x$add("surv.rmse", MeasureSurvRMSE)
	x$add("surv.mse", MeasureSurvMSE)
	x$add("surv.mae", MeasureSurvMAE)
	}
	register_mlr3pipelines = function() {
	mlr3pipelines::add_class_hierarchy_cache(c("PredictionSurv", "Prediction"))
	x = utils::getFromNamespace("mlr_pipeops", ns = "mlr3pipelines")
	x$add("distrcompose", PipeOpDistrCompositor)
	x$add("crankcompose", PipeOpCrankCompositor)
	x$add("breslowcompose", PipeOpBreslow, list(R6Class("Learner",
	public = list(id = "breslowcompose", task_type = "surv", predict_types = "lp",
	packages = c("mlr3", "mlr3proba"), param_set = ps()))$new()))
	x$add("trafotask_regrsurv", PipeOpTaskRegrSurv)
	x$add("trafotask_survregr", PipeOpTaskSurvRegr)
	x$add("trafopred_regrsurv", PipeOpPredRegrSurv)
	x$add("trafopred_survregr", PipeOpPredSurvRegr)
	x$add("compose_distr", PipeOpDistrCompositor)
	x$add("compose_crank", PipeOpCrankCompositor)
	x$add("compose_probregr", PipeOpProbregr)
	x$add("survavg", PipeOpSurvAvg)
	x = utils::getFromNamespace("mlr_graphs", ns = "mlr3pipelines")
	x$add("distrcompositor", pipeline_distrcompositor)
	x$add("crankcompositor", pipeline_crankcompositor)
	x$add("probregr", pipeline_probregr)
	x$add("survaverager", pipeline_survaverager)
	x$add("survbagging", pipeline_survbagging)
	x$add("survtoregr", pipeline_survtoregr)
	}

inside .onUnload below it.

Somewhat cumbersome and unfortunately not easily automatable as far as I can tell, maybe it would be even worth considering some generalized machinery for this in mlr3misc maybe?

[bblodfon]

Yep, we can definitely compress all this for sure, asked @sebffischer for help