How to use different dataset every epoch?

My mentor wants me to do the image novelty detection by normal two classes classification.the positive dataset is around 800 and the negative dataset genereted by a classification inference result for 2 millions.So I want to do generate a dataset which is combined by all the 800 positive dataset and 800 sampled from negative dataset with heavy data augmentation every epoch . I know there is randomsampler function in pytorch but it seems can’t sample only one class randomly.