I am learning the tutorial SVI part 1https://pyro.ai/examples/svi_part_i.html and I have some question about the example of coin.

In this example, which parameters are θ and which are φ(I guess these are ‘beta_q’ and ‘alpha_q’) in the expression of ELBO？Also,I noticed that in the model class there is no parameters assign to pyro,so does it means that there is no trainable θ?If that’s the case, the log probability will not increase forever.

This is my first time to ask for help in this forum and I would be happy if anyone could give me a hand

here theta is the empty set `{}`

, i.e. the model has latent variables but no learnable parameters. and yes here `phi = {alpha_q, beta_q}`

. and yes the marginal log probability of the model is fixed here (it depends on the data but not on any `theta`

) and so the goal of ELBO optimization is to fit `phi`

(there is no `theta)`

.

Thank you for your explain!I believe that this will play an important role in my future learning of SVI