Get ROC matrix

[ja-thomas]

nicer printer for the ROC matrix.

> print(x)
    predicted
true M         R                            
   M 14        15        TPR: 0.48 FNR: 0.52
   R 4         42        FPR: 0.09 TNR: 0.91
     PPV: 0.60 FOR: 0.26 LRP: 5.55 ACC: 0.75
     FDR: 0.40 NPV: 0.74 LRM: 0.57 DOR: 9.80


Abbreviations:
TPR - True positive rate (Sensisivity, Recall)
FPR - False positve rate (Fall-out)
FNR - False negative rate (Miss rate)
TNR - True negative rate (Specifity)
PPV - Positive predictive value (Precision)
FOR - False omission rate
LRP - Positive likelihood ratio (LR+)
ACC - Accuracy
LRM - Nevative likelihood ratio (LR-)
DOR - Diagnostic odds ratio

[berndbischl]

1. pls doc
2. the printer should have a flag to display the abbrevs
3. getROCMatrix with "M"
4. does relative = FALSE even make sense? remove?

[bhvieira]

Is ROC matrix a good name though?

[berndbischl]

Is ROC matrix a good name though?

well it was what came to our mind first.
it is basically this here
https://en.wikipedia.org/wiki/Receiver_operating_characteristic

better suggestion welcome

[PhilippPro]

confusion matrix?

at first glance roc matrix seemed to me a bit confusing (I thought about a matrix with the values of the ROC Curve)

[ja-thomas]

there is already getConfusionMatrix, which is also possible for multiclass problems.

Maybe it makes sense to combine both functions?

[berndbischl]

confusion matrix we already have.
calcROCMeasures?

reasons:

• it is no getter
• it returns no matrix
• clearer

we should then maybe also rename getConfMarix --> calcConfusionMatrix

[berndbischl]

and deprecate getConfMatrix

[berndbischl]

Maybe it makes sense to combine both functions?

hmmm. maybe. yes. probably

[ja-thomas]

should hopefully be good now. The name is calculateROCMeasures

[berndbischl]

1. pls rename als measures to lower case

2. pls doc the return type in a complete fashion. and conf.matrix not confMatrix.

3. for all defaults add them to the docs

4. use families "roc" and "performance"

[ja-thomas]

so the renaming for the measures is kind of annoying,

for example False Omission Rate -> for. I can't think of a diffferent three letter abbreviation. Everything else should be fixed.

[berndbischl]

@ja-thomas
i guess you need to fix this