Hi, so I want to plot a mean ROC curve in a multiclass setting averaged over several runs. Plotting works, but I have difficulties understanding whether I am plotting the ROC curve of an ensemble or the mean ROC of several runs. - I want to plot the mean ROC.

So let’s assume I run everything 5 times and I afterwards sum up the softmax values for every validation set example and divide the scores for every validation set example by 5.

If I then use the scikit learn function for every single class repeated like this for example:

`roc_curve(binarized_labels_class_x, softmax_probs_class_x)`

Does this give me the mean ROC curve for every class or is this some ensemble ROC curve? - Does an ensemble ROC curve actually exist?