categorical discriminator loss compared with 1

Adversarial_Autoencoder/semi_supervised_adversarial_autoencoder.py

Lines 235 to 236 in 2680a2b

    
           dc_c_loss_fake = tf.reduce_mean( 
        
               tf.nn.sigmoid_cross_entropy_with_logits(labels=tf.ones_like(d_c_fake), logits=d_c_fake))

Hi, I'm trying to implement the adversarial autoencoder myself. I found this repo and use it as a guide. It trains the autoencoder part but I think this line is a bug? Shouldn't the categorical discriminator train like the gaussian one? If it's not a bug could you comment on the training objective? Thanks!

Hey!

That is indeed an error from my part, I have fixed it.
Thank you for pointing it out!

	dc_c_loss_fake = tf.reduce_mean(
	tf.nn.sigmoid_cross_entropy_with_logits(labels=tf.ones_like(d_c_fake), logits=d_c_fake))