Compare ROCs - alternative for polynominal labels?

frank_claessen
frank_claessen New Altair Community Member
edited November 5 in Community Q&A
Hi all,

This is my first question here. 

Mind you, it is not necessarily forcing a roc-chart on a polynominal label but just running multiple models on the same dataset and after that finding some way of comparing accuracy, recall etc. from the respective applied models.

And - the AI I just consulted says there is no operator to apply more than one model like the way ROCS does. You just have to work around it by:

<citing gemini now>
"
  1. Convert Labels:
  • Use the "Nominal to Binominal" operator. This simplifies your problem to a binary classification by combining multiple label classes into two. However, this approach loses information about the original class distribution.
  1. One-vs-Rest Approach:
  • Build separate models for each unique class in the polynomial label. Each model predicts that class as positive and all others as negative.
  • Use "Compare ROCs" on these individual models. This provides a more comprehensive view of model performance for each class.
"
I hope to find people "smarter" than gemini

TIA - Frank
Tagged: