Open medcx opened 1 week ago
Hello,
What you mentioned refers to the "Ours_best" option in the paper, and you're right—it isn't a true validation (test). The code is set up so that you can test using other options, like "Ours_avg," which only uses the style key from the training set.
The style key of the validation set should be unknown, but you generate it directly during data preprocessing and use it directly in the sampling stage. Is this reasonable?