search

markhliu / DGAI

Learn Generative AI with PyTorch (Manning Publications, 2024)
MIT License
18 stars 14 forks source link

Question about first GAN in book #2

Open m7mdhka opened 3 days ago

m7mdhka commented 3 days ago

Hello, I want to ask about the first GAN we have trained on the book, Is it better to train the discriminator in both fake and real? I mean is to merge the fake samples with real samples in the same data loader, and then shuffle them and train the discriminator after. If no, why we train the discriminator on real first, then on fake samples?

markhliu commented 3 days ago

It doesn't matter. Merging real and fake samples when training won't help. You'll get the same result.