Greetings, I have problem understanding how ROC exactly imply. So from what I understood through the various videos I have gone through is that all classification machine learning predict by probability as per the function "predict_proba" which shows you the negative and positive probabilities and by default, the class with > 0.5 probability becomes positive. But we might not want that in some cases where we want our model to be more confident for example we might want to pick only > 0.75 or so that is where ROC is supposed to come to play to show the performance through various thresholds 0.0 to 1.0
#What do ROC and AUC imply?
worn estuary
But here is what I dont get, seems like we can make up the ROC independently of the model and each algorithm has different behavior so how are we relying on only TRP and FPR to decide about a model?
if you dont exactly understand what i am saying, please feel free to just explain the concepts in another way. I might be messing everything up I know. 
inner topaz
But here is what I dont get, seems like we can make up the ROC independently of ...
I don't exactly understand your question, but what I would say is that the ROC is not independent of the model. Different logistic regression models can have completely different TPR and FPR at different decision thresholds or even the same decision thresholds.
worn estuary
I don't exactly understand your question, but what I would say is that the ROC i...
thank you very much, i think im kind of getting it. i am still somewhat confused how TPR and FPR alone can be determinants for an algorithm but for now i will take it as is. there are some things that you understand overtime i believe as they are related to other things you might not know yet.