Why only use w as the input of Pz_wy_graph in GMVAE_graph.py

[MrShellx-1999820]

In GMVAE, p(x,z,w,y)=p(w)p(y)p(z|w,y)p(x|z). As you mentionend that in the generating process, z is generated from p(z|y, w) which means we first need to sample from p(y) and p(w). But in your code (GMVAE_graph.py), the input of the function Pz_wy_graph only contains w, where is y?

[psanch21]

Hi @MrShellx-1999820,

Thanks for your interest in my work! Notice that the distribution over z is define p(z|w,y)= N(mu_y(w), sigma_y(w)). You can see both parameters (mean and variance) are a function of w and y. InPz_wy_graph it is implemented in the following way. First, we have h = DenseNet(w) (lines 56-68), this part indeed only depends on w. Then, we have two for loops that will output the means and variances for each value of y. I.e., mean_k = DenseNet_k(h) and sigma_k = DenseNet_k(h) for k=1...K. Depending on the value of y, we will select one out of the K sets of parameters for p(z|w,y).

I hope this is helpful!