parameter tuning output NA

Question

I am new to parameter tuning with mlr package. I recently tried it with xgboost algorithm on a binary classification problem. I couldn’t get the trained accuracy, only NA. After google round, I was not able to debug my code. Could you give me some advice please?

Here is a reproducible example using mtcars data in R base:

    library(mlr)
    library(parallelMap)
    mtcars
    set.seed(1)
    train.index=sample(nrow(mtcars),nrow(mtcars)*0.7)
    train=mtcars[train.index,]
    test=mtcars[-train.index,]
    list(dim(train),dim(test))

    # set the tuning
    tune.dat=train[,c('mpg','cyl','disp','wt','vs','am')]
    traintask=makeClassifTask(data=tune.dat,target='am')
    learner=makeLearner('classif.xgboost',predict.type='response',nrounds=300,nthread=2)
    prange=makeParamSet(
    makeNumericParam('eta',lower=0.01,upper=0.3),
    makeNumericParam('max_depth',lower=2,upper=10),
    makeNumericParam('subsample',lower=0.4,upper=0.8),
    makeNumericParam('colsample_bytree',lower=0.4,upper=0.8)
    )
    ctrl=makeTuneControlRandom(maxit=50)
    rdesc=makeResampleDesc('CV',iters=4)
    parallelStartMulticore(2)

    # tune
    mytune=tuneParams(learner=learner,task=traintask,par.set=prange,control=ctrl,resampling=rdesc)
    mytune$y

the result should be a number rather than NA

Answer 1

I found 2 issues with your code:

1)The 'am' field wants to be a factor

tune.dat[, 'am'] <- as.factor(tune.dat[, 'am'])

2)The max_depth parameter wants to be an integer

makeIntegerParam('max_depth', lower = 2, upper = 10),

With these substitutions the output is

mmce.test.mean 
     0.225