I have built a deep learning model, where it does training for data with fixed noise level (which is low) after the deep learning model finishes the training, I start doing test for it by data with different levels of noise. For example, the level of data/noise during the training is 0.7 which means the noise power is weak, and then the testing stage the noise power will be varied which means the level of data/noise will go from 0:0.1:1.
What I need to know, when designing the ROC graph, should I design is after I finish the training stage at the same point of training (data/noise = 0.7) or I should do it for testing stage? if it should be done for test stage, in which level of data/noise I should do it?