I have a doubt regarding the above in the unsupervised setting. In src/trainer.py, on line number 77 and 78 , the labels assigned to the mapped vectors and real target vectors are 0.9 and 0.1 respectively (assuming discriminator smoothing has the default value of 0.1). Shouldn't that be opposite (as the discriminator should learn to label mapped vectors as fake (=0) and the target vectors as true (= 1)).
I have a doubt regarding the above in the unsupervised setting. In src/trainer.py, on line number 77 and 78 , the labels assigned to the mapped vectors and real target vectors are 0.9 and 0.1 respectively (assuming discriminator smoothing has the default value of 0.1). Shouldn't that be opposite (as the discriminator should learn to label mapped vectors as fake (=0) and the target vectors as true (= 1)).