I trained and tested an image classifier twice on the same dataset (with a proper train-val-test split, of course) with 20 classes, and saved the classification report. However, I noticed that the support looks like a fixed series of numbers, and not based on the actual number of test images per class. Classification report of the first test: '''''' Classification report of the second test: Note the the support in both tests are in the exactly same order, will class are not. Does anyone know the reason behind this? Thank you in advance! Continue reading...
