Skip to content

Question about calculate_kl #73

Description

@longcwang3

For calculate_kl, I am assuming mu_q refers to the variational posterior and mu_p refers to the prior.

def calculate_kl(mu_q, sig_q, mu_p, sig_p):
kl = 0.5 * (2 * torch.log(sig_p / sig_q) - 1 + (sig_q / sig_p).pow(2) + ((mu_p - mu_q) / sig_p).pow(2)).sum()
return kl

While in the following codes, prior_mu is input as mu_q.

def kl_loss(self):
kl = KL_DIV(self.prior_mu, self.prior_sigma, self.W_mu, self.W_sigma)
if self.use_bias:
kl += KL_DIV(self.prior_mu, self.prior_sigma, self.bias_mu, self.bias_sigma)
return kl

Could you check this?

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions