Why the n_samples in SVI is DEPRECATED?

Misc.
1

Hi,

I am trying to understand why this parameter n_samples in SVI is deprecated, I guess with more Monte-Carlo samples the noisy estimates for gradients would be more stable or accurate. Did any paper/research suggest one Monte-Carlo sample is better? Thanks.

2

are you referring to num_particles? it is not deprecated. see e.g. this example