question about num_aug

[RenShuhuai-Andy]

Hi, I think here in augmentation_main.py

learning-data-manipulation/augmentation_main.py

Lines 99 to 100 in 315313d

	batch_examples = generator.augment_batch(
	examples=batch_examples, num_aug=1)

num_aug should be set to args.n_aug instead of 1. Is it a typo? Or you only generate args.n_aug augmented texts for fine-tuning generator? But why?

[tanyuqian]

No, it's not a typo. We augment args.num_aug examples to finetune the generator (Line97) and augment 1 example to train the classifier (Line100 & Line102). They are two separated hyperparameters.