CODEWITHSUNDEEP
neural-networkroc
nick_green
nick_green

Reputation: 1664

How to draw ROC curve from given data

I know it's stupid question and I know it was occurred a lot in stack overflow, but I can't understand any of examples...

for example I have data that I got from my neural network. I know it's not much data, but I hope it will be enough to show me an example...

Speaker | Guessed speaker
_________________________
1       |       2
2       |       2
3       |       3
4       |       9
5       |       5
6       |       6
7       |       7
8       |       1
9       |       9
10      |       10

So now, how should I create ROC curve from this data?

Upvotes: 0

Views: 523

Answers (1)

Dr. Snoopy
Dr. Snoopy

Reputation: 56377

You cannot make a ROC curve from this data, ROC curves are made from scores that can be thresholded to produce actual predictions, like for example, confidence scores. You need to produce these scores. If your classifier outputs softmax probabilities, then that can be used.

An ROC curve is produced by setting a threshold on the score and producing predictions, then you can compute true and false positives, and produce one point on the curve. You change the threshold, and another point is produced, and so on.

Also note that ROC curves are mainly for binary classification problems, and can be extended to multi-class problems by making one curve per class, using class vs non-class as a binary decision.

Upvotes: 1

Related Questions