How to interpret the batch accuracy for train and test
opened this issue · comments
Deleted user commented
Hiyam Ghannam commented
Hello, this is due to the fact that, according to the paper, we sample K + 1 datapoints, We train on the K data points and evaluate only on the last one, therefore, the accuracy is described as num_correct / num_classes
so basically if you have 2 classes and you predicted 1 correct and the other not, then the accuracy = 1 / 2 = 0.5