What is the TF equivalent for the "obs" argument in Torch pyro.sample()?

TensorFlow General Discussion
,
1

am stuck on this question trying to convert a VAE in PyTorch/Pyro to TFP/Edward2. The original Code can be found here.

So the basic question is how to convert these code snippet to TFP/Edward2.

with pyro.plate("data", x.size(0)):
    t = pyro.sample("t", self.t_dist(z), obs=t)

with pyro.plate("data", size, subsample=x):
    t = pyro.sample("t", self.t_dist(z), obs=t, infer={"is_auxiliary": True})

The t is of shape t.shape = [673, 1] , the x of shape x.shape = [673, 25] and the self.t_dist is a Bernoulli Distribution Network. z is sampled from a normal distributionz = pyro.sample("z", self.z_dist()) with:

def z_dist(self):
     return dist.Normal(0, 1).expand([15]).to_event(1)

Thanks for the Help!!