library(mlr)
library(tidyverse)
library(reshape2)
data("airquality")
air=airquality %>%
filter(!is.na(Ozone))
lin1=ggplot(air) +
aes(
y = Solar.R,
x = Ozone,
colour = Ozone,
size = Ozone,
) +
geom_point(alpha=0.6,size=6) +
geom_smooth(color="red",se=TRUE)+
scale_color_viridis_c(option = "magma", direction = -1) +
theme_bw()
lin2=ggplot(air) +
aes(
y = Temp,
x = Ozone,
colour = Ozone,
size = Ozone,
) +
geom_point(alpha=0.6,size=6) +
geom_smooth(color="red",se=TRUE)+
scale_color_viridis_c(option = "magma", direction = -1) +
theme_bw()
lin3=ggplot(air) +
aes(
y = Wind,
x = Ozone,
colour = Ozone,
size = Ozone,
) +
geom_point(alpha=0.6,size=6) +
geom_smooth(color="red",se=TRUE)+
scale_color_viridis_c(option = "magma", direction = -1) +
theme_bw()
lin4=ggplot(air) +
aes(
y = Month,
x = Ozone,
colour = Ozone,
size = Ozone,
) +
geom_point(alpha=0.6,size=6) +
geom_smooth(color="red",se=TRUE)+
scale_color_viridis_c(option = "magma", direction = -1) +
theme_bw()
lin5=ggplot(air) +
aes(
y = Day,
x = Ozone,
colour = Ozone,
size = Ozone,
) +
geom_point(alpha=0.6,size=6) +
geom_smooth(color="red",se=TRUE)+
scale_color_viridis_c(option = "magma", direction = -1) +
theme_bw()
library(patchwork)
(lin1+lin2+lin3)/(lin4+lin5)
fit=lm(Ozone~.,data=air)
summary(fit)
##
## Call:
## lm(formula = Ozone ~ ., data = air)
##
## Residuals:
## Min 1Q Median 3Q Max
## -37.014 -12.284 -3.302 8.454 95.348
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -64.11632 23.48249 -2.730 0.00742 **
## Solar.R 0.05027 0.02342 2.147 0.03411 *
## Wind -3.31844 0.64451 -5.149 1.23e-06 ***
## Temp 1.89579 0.27389 6.922 3.66e-10 ***
## Month -3.03996 1.51346 -2.009 0.04714 *
## Day 0.27388 0.22967 1.192 0.23576
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 20.86 on 105 degrees of freedom
## (5 observations deleted due to missingness)
## Multiple R-squared: 0.6249, Adjusted R-squared: 0.6071
## F-statistic: 34.99 on 5 and 105 DF, p-value: < 2.2e-16
library(naniar)
vis_miss(air,cluster = F)
visdat::vis_dat(air,palette = "cb_safe" )
task=makeRegrTask(data=air,target = "Ozone")
lrn.lm=makeLearner("regr.lm")
lrn.gam=makeLearner("regr.gamboost")
#if our model over fits then we can use below models
#if does not we will do it anyways " _ "
lrn.ridge=makeLearner("regr.glmnet",alpha=0,id="ridge")
lrn.lasso=makeLearner("regr.glmnet",alpha=1,id="Lasso")
lrn.elastic=makeLearner("regr.glmnet",id="elastic")
#linearmodel
imputeMethod=imputeLearner("regr.rpart")
lrn.lm.imp=makeImputeWrapper(learner = lrn.lm, classes = list(integer=imputeMethod) )
#filteringFeature
fval=generateFilterValuesData(task,method = "linear.correlation")
plotFilterValues(fval)
lrn.lm.imp.filter=makeFilterWrapper(learner = lrn.lm.imp,fw.method = "linear.correlation")
getParamSet(lrn.lm.imp.filter)
## Type len Def Constr
## fw.method discrete - - anova.test,auc,carscore,cforest.impor...
## fw.base.methods discrete - - anova.test,auc,carscore,cforest.impor...
## fw.perc numeric - - 0 to 1
## fw.abs integer - - 0 to Inf
## fw.threshold numeric - - -Inf to Inf
## fw.fun function - - -
## fw.fun.args untyped - <NULL> -
## fw.mandatory.feat untyped - - -
## tol numeric - 1e-07 0 to Inf
## singular.ok logical - TRUE -
## Req Tunable Trafo
## fw.method - TRUE -
## fw.base.methods - TRUE -
## fw.perc - TRUE -
## fw.abs - TRUE -
## fw.threshold - TRUE -
## fw.fun - TRUE -
## fw.fun.args - TRUE -
## fw.mandatory.feat - TRUE -
## tol - TRUE -
## singular.ok - FALSE -
#parameters set
ps=makeParamSet(
makeNumericParam("fw.perc",0.3,1)
)
sc=makeTuneControlGrid()
kfold=makeResampleDesc("CV", iters=5)
#crossvalidation and performace
tp=tuneParams(learner = lrn.lm.imp.filter,task = task , par.set = ps, resampling = kfold, control = sc,rmse)
#or
lrn.lm.imp.filter.tune=makeTuneWrapper(learner = lrn.lm.imp.filter, par.set = ps, resampling = kfold, control = sc)
CV=resample(learner = lrn.lm.imp.filter.tune,task = task ,resampling = kfold,rmse)
model1=train(lrn.lm.imp.filter.tune,task)
pred1=predict(model1,task)
tp
## Tune result:
## Op. pars: fw.perc=0.533
## rmse.test.rmse=21.6343771
## Resample Result
## Task: air
## Learner: regr.lm.imputed.filtered.tuned
## Aggr perf: rmse.test.rmse=22.0180357
## Runtime: 6.31697
## rmse
## 20.67302
## Model for learner.id=regr.lm.imputed.filtered; learner.class=FilterWrapper
## Trained on: task.id = air; obs = 116; features = 5
## Hyperparameters: fw.perc=0.767
#nonlinear model
task=makeRegrTask(data=air,target = "Ozone")
lrn.gamboost=makeLearner("regr.gamboost")
#impute
imputeMethod=imputeLearner("regr.rpart")
lrn.gam.impute=makeImputeWrapper(learner = lrn.gam, classes = list(integer=imputeMethod))
#automate filter feature
lrn.gam.impute.filter=makeFilterWrapper(learner = lrn.gam.impute,fw.method = "linear.correlation")
getParamSet(lrn.gam.impute.filter)
## Type len Def
## fw.method discrete - -
## fw.base.methods discrete - -
## fw.perc numeric - -
## fw.abs integer - -
## fw.threshold numeric - -
## fw.fun function - -
## fw.fun.args untyped - <NULL>
## fw.mandatory.feat untyped - -
## baselearner discrete - -
## dfbase integer - 4
## offset numeric - -
## family discrete - Gaussian
## custom.family.definition untyped - -
## nuirange numericvector <NA> 0,100
## d numeric - -
## mstop integer - 100
## nu numeric - 0.1
## risk discrete - -
## stopintern logical - FALSE
## trace logical - FALSE
## Constr Req Tunable
## fw.method anova.test,auc,carscore,cforest.impor... - TRUE
## fw.base.methods anova.test,auc,carscore,cforest.impor... - TRUE
## fw.perc 0 to 1 - TRUE
## fw.abs 0 to Inf - TRUE
## fw.threshold -Inf to Inf - TRUE
## fw.fun - - TRUE
## fw.fun.args - - TRUE
## fw.mandatory.feat - - TRUE
## baselearner bbs,bols,btree - TRUE
## dfbase -Inf to Inf - TRUE
## offset -Inf to Inf - TRUE
## family Gaussian,Laplace,Huber,Poisson,GammaR... - TRUE
## custom.family.definition - Y TRUE
## nuirange -Inf to Inf Y TRUE
## d -Inf to Inf Y TRUE
## mstop 1 to Inf - TRUE
## nu 0 to 1 - TRUE
## risk inbag,oobag,none - TRUE
## stopintern - - TRUE
## trace - - FALSE
## Trafo
## fw.method -
## fw.base.methods -
## fw.perc -
## fw.abs -
## fw.threshold -
## fw.fun -
## fw.fun.args -
## fw.mandatory.feat -
## baselearner -
## dfbase -
## offset -
## family -
## custom.family.definition -
## nuirange -
## d -
## mstop -
## nu -
## risk -
## stopintern -
## trace -
#paraset
ps=makeParamSet(
makeNumericParam("fw.perc",0.2,1)
)
kfold=makeResampleDesc("CV",iters=5)
tc=makeTuneControlGrid()
#parameter tuning and cross vaalidatiom
tp=tuneParams(learner = lrn.gam.impute.filter,task = task,par.set = ps,resampling = kfold,control = tc,rmse)
#or
lrn.gam.impute.filter.tune=makeTuneWrapper(learner = lrn.gam.impute.filter,par.set = ps,resampling = kfold,control = tc)
resample(learner = lrn.gam.impute.filter.tune,task=task,resampling = kfold,rmse)
## Resample Result
## Task: air
## Learner: regr.gamboost.imputed.filtered.tuned
## Aggr perf: rmse.test.rmse=19.3045404
## Runtime: 13.3062
#model.train
model2=train(lrn.gam.impute.filter.tune,task)
prediction2=predict(model2,task)
performance(prediction2,rmse)
## rmse
## 17.02491
tuning=generateHyperParsEffectData(tune.result = tp)
tuning$data
## fw.perc rmse.test.rmse iteration exec.time
## 1 0.2000000 22.34807 1 0.17
## 2 0.2888889 22.34807 2 0.14
## 3 0.3777778 19.70743 3 0.19
## 4 0.4666667 19.70743 4 0.18
## 5 0.5555556 18.80696 5 0.30
## 6 0.6444444 18.80696 6 0.28
## 7 0.7333333 18.91867 7 0.33
## 8 0.8222222 18.91867 8 0.34
## 9 0.9111111 19.09023 9 0.39
## 10 1.0000000 19.09023 10 0.41
library(plotly)
plot=plotHyperParsEffect(tuning, y ="rmse.test.rmse",x="fw.perc",plot.type="line")
plot
#elastic net
#lasso
#ridge regression
task
## Supervised task: air
## Type: regr
## Target: Ozone
## Observations: 116
## Features:
## numerics factors ordered functionals
## 5 0 0 0
## Missings: TRUE
## Has weights: FALSE
## Has blocking: FALSE
## Has coordinates: FALSE
lrn.ridge=makeLearner("regr.glmnet", id="ridge",alpha=0)
imputeMethod=imputeLearner("regr.rpart")
lrn.ridge.imp=makeImputeWrapper(lrn.ridge,classes = list(integer=imputeMethod))
ps2=makeParamSet(makeNumericParam("s",0,100))
kfold2=makeResampleDesc("CV", iters=5)
tc=makeTuneControlGrid()
lrn.ridge.imp.tune=makeTuneWrapper(learner = lrn.ridge.imp,par.set=ps2,control = tc,resampling=kfold2)
crossvalidate=resample(learner = lrn.ridge.imp.tune,task = task, resampling = kfold2,rmse)
tp.ridge=tuneParams(learner = lrn.ridge.imp,par.set=ps2,control = tc,resampling=kfold2,task = task,rmse)
crossvalidate
## Resample Result
## Task: air
## Learner: ridge.imputed.tuned
## Aggr perf: rmse.test.rmse=21.6075845
## Runtime: 7.52145
ridge.hyperpars=generateHyperParsEffectData(tp.ridge)
plotHyperParsEffect(ridge.hyperpars,x="s",y="rmse.test.rmse",plot.type = "line")
model.ridge=train(lrn.ridge.imp.tune,task)
predict.ridge=predict(model.ridge,task)
performance(predict.ridge,rmse)
## rmse
## 20.51208
## Supervised task: air
## Type: regr
## Target: Ozone
## Observations: 116
## Features:
## numerics factors ordered functionals
## 5 0 0 0
## Missings: TRUE
## Has weights: FALSE
## Has blocking: FALSE
## Has coordinates: FALSE
lrn.lasso=makeLearner("regr.glmnet", id="lasso",alpha=1)
imputeMethod=imputeLearner("regr.rpart")
lrn.lasso.imp=makeImputeWrapper(lrn.lasso,classes = list(integer=imputeMethod))
ps2=makeParamSet(makeNumericParam("s",0,100))
kfold2=makeResampleDesc("CV", iters=10)
tc=makeTuneControlGrid()
lrn.lasso.imp.tune=makeTuneWrapper(learner = lrn.lasso.imp,par.set=ps2,control = tc,resampling=kfold2)
crossvalidate2=resample(learner = lrn.lasso.imp.tune,task = task, resampling = kfold2,rmse)
tp.lasso=tuneParams(learner = lrn.lasso.imp,par.set=ps2,control = tc,resampling=kfold2,task = task,rmse)
crossvalidate2
## Resample Result
## Task: air
## Learner: lasso.imputed.tuned
## Aggr perf: rmse.test.rmse=21.4163095
## Runtime: 27.6723
lasso.hyperpars=generateHyperParsEffectData(tp.lasso)
plotHyperParsEffect(lasso.hyperpars,x="s",y="rmse.test.rmse",plot.type = "line")
model.lasso=train(lrn.lasso.imp.tune,task)
predict.lasso=predict(model.lasso,task)
performance(predict.lasso,rmse)
## rmse
## 20.46227
## Supervised task: air
## Type: regr
## Target: Ozone
## Observations: 116
## Features:
## numerics factors ordered functionals
## 5 0 0 0
## Missings: TRUE
## Has weights: FALSE
## Has blocking: FALSE
## Has coordinates: FALSE
lrn.elasticnet=makeLearner("regr.glmnet",id="elastic")
imputeMethod=imputeLearner("regr.rpart")
lrn.elasticnet.impute=makeImputeWrapper(learner = lrn.elasticnet, classes = list(integer=imputeMethod,numeric=imputeMethod))
getParamSet(lrn.elasticnet.impute)
## Type len Def Constr Req Tunable
## family discrete - gaussian gaussian,poisson - TRUE
## alpha numeric - 1 0 to 1 - TRUE
## s numeric - - 0 to Inf - TRUE
## exact logical - FALSE - - TRUE
## nlambda integer - 100 1 to Inf - TRUE
## lambda.min.ratio numeric - - 0 to 1 - TRUE
## lambda numericvector <NA> - 0 to Inf - TRUE
## standardize logical - TRUE - - TRUE
## intercept logical - TRUE - - TRUE
## thresh numeric - 1e-07 0 to Inf - TRUE
## dfmax integer - - 0 to Inf - TRUE
## pmax integer - - 0 to Inf - TRUE
## exclude integervector <NA> - 1 to Inf - TRUE
## penalty.factor numericvector <NA> - 0 to 1 - TRUE
## lower.limits numericvector <NA> - -Inf to 0 - TRUE
## upper.limits numericvector <NA> - 0 to Inf - TRUE
## maxit integer - 100000 1 to Inf - TRUE
## standardize.response logical - FALSE - - TRUE
## fdev numeric - 1e-05 0 to 1 - TRUE
## devmax numeric - 0.999 0 to 1 - TRUE
## eps numeric - 1e-06 0 to 1 - TRUE
## big numeric - 9.9e+35 -Inf to Inf - TRUE
## mnlam integer - 5 1 to Inf - TRUE
## pmin numeric - 1e-09 0 to 1 - TRUE
## exmx numeric - 250 -Inf to Inf - TRUE
## prec numeric - 1e-10 -Inf to Inf - TRUE
## mxit integer - 100 1 to Inf - TRUE
## offset untyped - <NULL> - - TRUE
## type.gaussian discrete - - covariance,naive Y TRUE
## relax logical - FALSE - - TRUE
## Trafo
## family -
## alpha -
## s -
## exact -
## nlambda -
## lambda.min.ratio -
## lambda -
## standardize -
## intercept -
## thresh -
## dfmax -
## pmax -
## exclude -
## penalty.factor -
## lower.limits -
## upper.limits -
## maxit -
## standardize.response -
## fdev -
## devmax -
## eps -
## big -
## mnlam -
## pmin -
## exmx -
## prec -
## mxit -
## offset -
## type.gaussian -
## relax -
ps2=makeParamSet(makeNumericParam("s",0,100))
kfold3=makeResampleDesc("CV",iters=10)
tc=makeTuneControlGrid()
lrn.elasticnet.impute.tune=makeTuneWrapper(learner = lrn.elasticnet.impute,par.set = ps2,control = tc,resampling = kfold3)
cv=resample(task,learner=lrn.elasticnet.impute.tune,resampling = kfold3 ,rmse)
tp.elasticnet=tuneParams(task,learner=lrn.elasticnet.impute.tune,resampling = kfold3 ,par.set = ps2,control = tc,rmse)
model.elastic=train(learner=lrn.elasticnet.impute.tune,task)
elastic.predict=predict(model.elastic,task)
performance(elastic.predict,rmse)
## rmse
## 20.46227
## Resample Result
## Task: air
## Learner: elastic.imputed.tuned
## Aggr perf: rmse.test.rmse=21.8699670
## Runtime: 31.1308
## Supervised task: air
## Type: regr
## Target: Ozone
## Observations: 116
## Features:
## numerics factors ordered functionals
## 5 0 0 0
## Missings: TRUE
## Has weights: FALSE
## Has blocking: FALSE
## Has coordinates: FALSE
lrn.knn=makeLearner("regr.kknn")
imputeMethod=imputeLearner("regr.rpart")
lrn.knn.imp=makeImputeWrapper(learner = lrn.knn,classes = list(numeric=imputeMethod,integer=imputeMethod))
getParamSet(lrn.knn.imp)
## Type len Def Constr Req
## k integer - 7 1 to Inf -
## distance numeric - 2 0 to Inf -
## kernel discrete - optimal rectangular,triangular,epanechnikov,b... -
## scale logical - TRUE - -
## Tunable Trafo
## k TRUE -
## distance TRUE -
## kernel TRUE -
## scale TRUE -
ps=makeParamSet(
makeIntegerParam("k",1,30)
)
kfold=makeResampleDesc("CV",iters=10)
sc=makeTuneControlGrid()
lrn.knn.imp.tune=makeTuneWrapper(learner=lrn.knn.imp,resampling=kfold,control=sc,par.set = ps)
cv=resample(learner = lrn.knn.imp.tune,task=task,resampling = kfold,rmse)
cv
## Resample Result
## Task: air
## Learner: regr.kknn.imputed.tuned
## Aggr perf: rmse.test.rmse=19.4788632
## Runtime: 29.2668
tp=tuneParams(learner=lrn.knn.imp,resampling=kfold,control=sc,par.set = ps,task=task,rmse)
model.kknn=train(learner = lrn.knn.imp.tune,task = task)
predict.knn=predict(model.kknn,task)
performance(predict.knn,rmse)
## rmse
## 12.7292
## Supervised task: air
## Type: regr
## Target: Ozone
## Observations: 116
## Features:
## numerics factors ordered functionals
## 5 0 0 0
## Missings: TRUE
## Has weights: FALSE
## Has blocking: FALSE
## Has coordinates: FALSE
lrn.rand=makeLearner("regr.randomForest")
impute=imputeLearner("regr.rpart")
lrn.rand.impute=makeImputeWrapper(learner = lrn.rand,classes = list(integer=impute))
getParamSet(lrn.rand.impute)
## Type len Def Constr Req Tunable Trafo
## ntree integer - 500 1 to Inf - TRUE -
## se.ntree integer - 100 1 to Inf Y TRUE -
## se.method discrete - sd bootstrap,jackknife,sd Y TRUE -
## se.boot integer - 50 1 to Inf - TRUE -
## mtry integer - - 1 to Inf - TRUE -
## replace logical - TRUE - - TRUE -
## strata untyped - - - - FALSE -
## sampsize integervector <NA> - 1 to Inf - TRUE -
## nodesize integer - 5 1 to Inf - TRUE -
## maxnodes integer - - 1 to Inf - TRUE -
## importance logical - FALSE - - TRUE -
## localImp logical - FALSE - - TRUE -
## nPerm integer - 1 -Inf to Inf - TRUE -
## proximity logical - FALSE - - FALSE -
## oob.prox logical - - - Y FALSE -
## do.trace logical - FALSE - - FALSE -
## keep.forest logical - TRUE - - FALSE -
## keep.inbag logical - FALSE - - FALSE -
ps=makeParamSet(
makeIntegerParam("ntree",30,30),
makeIntegerParam("mtry",1,30),
makeIntegerParam("nodesize",5,10),
makeIntegerParam("maxnodes",1,30)
)
kfold=makeResampleDesc("CV",iters=5)
sc=makeTuneControlRandom(maxit = 10)
lrn.rnd.impute.tune=makeTuneWrapper(learner = lrn.rand.impute,kfold,par.set = ps,control = sc)
resample(lrn.rnd.impute.tune,task,kfold,rmse)
## Resample Result
## Task: air
## Learner: regr.randomForest.imputed.tuned
## Aggr perf: rmse.test.rmse=18.9157454
## Runtime: 7.37146
## Supervised task: air
## Type: regr
## Target: Ozone
## Observations: 116
## Features:
## numerics factors ordered functionals
## 5 0 0 0
## Missings: TRUE
## Has weights: FALSE
## Has blocking: FALSE
## Has coordinates: FALSE
lrn.xg=makeLearner("regr.xgboost")
method=imputeLearner("regr.rpart")
lrn.xg.imp=makeImputeWrapper(learner = lrn.xg , classes=list(integer=method))
getParamSet(lrn.xg.imp)
## Type len Def
## booster discrete - gbtree
## watchlist untyped - <NULL>
## eta numeric - 0.3
## gamma numeric - 0
## max_depth integer - 6
## min_child_weight numeric - 1
## subsample numeric - 1
## colsample_bytree numeric - 1
## colsample_bylevel numeric - 1
## colsample_bynode numeric - 1
## num_parallel_tree integer - 1
## lambda numeric - 1
## lambda_bias numeric - 0
## alpha numeric - 0
## objective untyped - reg:squarede...
## eval_metric untyped - rmse
## base_score numeric - 0.5
## max_delta_step numeric - 0
## missing numeric -
## monotone_constraints integervector <NA> 0
## tweedie_variance_power numeric - 1.5
## nthread integer - -
## nrounds integer - -
## feval untyped - <NULL>
## verbose integer - 1
## print_every_n integer - 1
## early_stopping_rounds integer - <NULL>
## maximize logical - <NULL>
## sample_type discrete - uniform
## normalize_type discrete - tree
## rate_drop numeric - 0
## skip_drop numeric - 0
## scale_pos_weight numeric - 1
## refresh_leaf logical - TRUE
## feature_selector discrete - cyclic
## top_k integer - 0
## predictor discrete - cpu_predictor
## updater untyped - -
## sketch_eps numeric - 0.03
## one_drop logical - FALSE
## tree_method discrete - auto
## grow_policy discrete - depthwise
## max_leaves integer - 0
## max_bin integer - 256
## callbacks untyped - list()
## Constr Req Tunable Trafo
## booster gbtree,gblinear,dart - TRUE -
## watchlist - - FALSE -
## eta 0 to 1 - TRUE -
## gamma 0 to Inf - TRUE -
## max_depth 0 to Inf - TRUE -
## min_child_weight 0 to Inf - TRUE -
## subsample 0 to 1 - TRUE -
## colsample_bytree 0 to 1 - TRUE -
## colsample_bylevel 0 to 1 - TRUE -
## colsample_bynode 0 to 1 - TRUE -
## num_parallel_tree 1 to Inf - TRUE -
## lambda 0 to Inf - TRUE -
## lambda_bias 0 to Inf - TRUE -
## alpha 0 to Inf - TRUE -
## objective - - FALSE -
## eval_metric - - FALSE -
## base_score -Inf to Inf - FALSE -
## max_delta_step 0 to Inf - TRUE -
## missing -Inf to Inf - FALSE -
## monotone_constraints -1 to 1 - TRUE -
## tweedie_variance_power 1 to 2 Y TRUE -
## nthread 1 to Inf - FALSE -
## nrounds 1 to Inf - TRUE -
## feval - - FALSE -
## verbose 0 to 2 - FALSE -
## print_every_n 1 to Inf Y FALSE -
## early_stopping_rounds 1 to Inf - FALSE -
## maximize - - FALSE -
## sample_type uniform,weighted Y TRUE -
## normalize_type tree,forest Y TRUE -
## rate_drop 0 to 1 Y TRUE -
## skip_drop 0 to 1 Y TRUE -
## scale_pos_weight -Inf to Inf - TRUE -
## refresh_leaf - - TRUE -
## feature_selector cyclic,shuffle,random,greedy,thrifty - TRUE -
## top_k 0 to Inf - TRUE -
## predictor cpu_predictor,gpu_predictor - TRUE -
## updater - - TRUE -
## sketch_eps 0 to 1 - TRUE -
## one_drop - Y TRUE -
## tree_method auto,exact,approx,hist,gpu_hist Y TRUE -
## grow_policy depthwise,lossguide Y TRUE -
## max_leaves 0 to Inf Y TRUE -
## max_bin 2 to Inf Y TRUE -
## callbacks - - FALSE -
ps=makeParamSet(
makeNumericParam("eta",0.3,0.5),
makeNumericParam("gamma",0,0.5),
makeIntegerParam("max_depth",6,12),
makeNumericParam("min_child_weight",1,10),
makeNumericParam("subsample",0.5,1),
makeIntegerParam("nrounds",30,30),
makeNumericParam("colsample_bytree",0.5,1)
)
sc=makeTuneControlRandom(maxit = 10)
kfold=makeResampleDesc("CV",iters=10)
lrn.xg.imp.tune=makeTuneWrapper(learner = lrn.xg.imp,par.set = ps,control = sc,resampling = kfold)
resample(lrn.xg.imp.tune,task,rmse,resampling=kfold)
## Resample Result
## Task: air
## Learner: regr.xgboost.imputed.tuned
## Aggr perf: rmse.test.rmse=19.2146588
## Runtime: 71.9593