exercise_tuning.Rmd

# Exercise: Tuning

- Create a classification task for the  Ionosphere data set (package mlbench)
- Remove constant features from the data set.
- Fit a random forest on this task. Tune the parameters `mtry` and `nodesize`
  using a 3-fold cross validation and the random search (see ?makeTuneControlRandom).
  Set the budget via `maxit` to 20.
- Compare the mean misclassification error (mmce) with the results of single classification tree from the rpart package.



# Solution
```{r}
library(mlr)
data("Ionosphere", package = "mlbench")
task = makeClassifTask(data = Ionosphere, target = "Class", positive = "good")
task = removeConstantFeatures(task)
lrn = makeLearner("classif.randomForest")
rdesc = makeResampleDesc("CV", iters = 3)
rin = makeResampleInstance(rdesc, task)
ps = makeParamSet(
  makeIntegerParam("mtry", lower = 1, upper = 10),
  makeIntegerParam("nodesize", lower = 1, upper = 10)
)
ctrl = makeTuneControlRandom(maxit = 20)
res = tuneParams(learner = lrn, task = task, resampling = rin, measures = list(mmce), par.set = ps, control = ctrl)
print(res)
resample("classif.rpart", task = task, resampling = rin)
```