Exporting ROC Curve underlying data

malcolmhaynes

Is there anyway to export the underlying data used to create the ROC curve from the performance operator? I want a table where each row has a decision threshold, TPR, and FPR. 

MartinLiebig

Hi,

Converters extension has a operator called ROC Curve to Example Set which does this for you.

Best,

Martin

MartinLiebig

Hi,

Converters extension has a operator called ROC Curve to Example Set which does this for you.

Best,

Martin

malcolmhaynes

The converter operator has one parameter - number of evaluation points. How is the converter generating the evaluation points? For example, if I have a dataset with 250 examples, but the evaluation points are set to 500. Is it just extrapolating points from a best fit curve?

MartinLiebig

Hi @malcolmhaynes ,

been a while since I dived into this (when writing the AUPRC operator). The short answer is yes. The usual way how we implement a ROC curve is using an interpolation. The code for it is available here: https://github.com/rapidminer/rapidminer-studio/blob/master/src/main/java/com/rapidminer/tools/math/ROCData.java (and in other classes of the math package).

I hope this helps,

Martin