0

My softmax classifier outputs 312 output probabilities; each referring to an individual class (total classes = 312).
How can I show the results as a ROC curve?
Do I have to plot ROC for each class? (Doing this will creates 312 curves.)
Any way to show the ROC curve for such multi class classification problems?

greybeard
  • 2,249
  • 8
  • 30
  • 66
Lakpa Tamang
  • 408
  • 1
  • 4
  • 12
  • Related to how to compute such metric, probably [this](https://stackoverflow.com/questions/11424112/multiclass-roc-curves-in-r/65288174#65288174) and [this](https://stackoverflow.com/questions/36631054/roc-curves-for-multiclass-classification-in-r/65287938#65287938) could help you. For an extended discussion about the topic [this question in StackExchange](https://stats.stackexchange.com/questions/2151/how-to-plot-roc-curves-in-multiclass-classification) could be useful too – Álvaro A. Gutiérrez-Vargas Dec 29 '20 at 09:26
  • Appreciate your comment. However, i' m working this on python. Aren't there any built in libs for drawing the same in python. For classes as huge as 312 (output nodes), drawing onevsall does not seem feasible either. Is there any approach where i can show my multiclass classification results. (not necessasarily need to be the ROC-AUC). – Lakpa Tamang Dec 30 '20 at 02:35
  • this is the only resource I know that could be of your interest. https://scikit-learn.org/stable/auto_examples/model_selection/plot_roc.html – Álvaro A. Gutiérrez-Vargas Dec 30 '20 at 13:48
  • (If and when you think your question programming language specific, please follow tag `softmax`' prompt and tag with language.) – greybeard Dec 31 '20 at 08:31

0 Answers0