separate encoder output for code vectors

[vandana-rajan]

Hi,

I was going through main.py and I have a doubt regarding the 'network architecture section'.

`# network architecture

y, z, neg_marginal_likelihood, D_loss, G_loss = aae.adversarial_autoencoder(x_hat, x, x_id, z_sample, z_id, dim_features, dim_z, n_hidden, keep_prob)

z_train     = aae.encoder(x, n_hidden, dim_z)

z_test      = aae.encoder(x, n_hidden, dim_z)`

Why are you creating separate 'aae.encoder'? The code vectors are already given as output of 'aae.adversarial_autoencoder'.

Vandana Rajan

[eesungkim]

Hi,

It is a good point.
There is no need to write two. :)

Thanks!

[vandana-rajan]

Thanks for your reply.
I am not an expert in tensorflow and I am trying to understand your code.
When you created separate 'aae.encoder' and extracted code vectors, I assume that the already trained encoder part of the network is used because of the 'reuse=True' in the encoder function. Am I right?

[piaotuan]

你好，可以发一下你有关训练对抗自编码器时绘制损失函数图像的源码码