fixing docs

Author: liopeer

diffusion_models/losses/kl_divergence.py

@@ -11,6 +11,8 @@ def gaussian_kl(
     """Calculate KL Divergence of 2 Gaussian distributions.
 
     KL divergence between two univariate Gaussians, as derived in [1], with k=1 (dimensionality).
+    .. math::
+        D_{KL}(p||q) = \frac{1}{2}\left[\log\frac{|\Sigma_q|}{|\Sigma_p|} - k + (\boldsymbol{\mu_p}-\boldsymbol{\mu_q})^T\Sigma_q^{-1}(\boldsymbol{\mu_p}-\boldsymbol{\mu_q}) + tr\left\{\Sigma_q^{-1}\Sigma_p\right\}\right]
 
     Parameters
     ----------
@@ -30,11 +32,7 @@ def gaussian_kl(
 
     References
     ----------
-    .. math::
-        D_{KL}(p||q) = \frac{1}{2}\left[\log\frac{|\Sigma_q|}{|\Sigma_p|} - k + (\boldsymbol{\mu_p}-\boldsymbol{\mu_q})^T\Sigma_q^{-1}(\boldsymbol{\mu_p}-\boldsymbol{\mu_q}) + tr\left\{\Sigma_q^{-1}\Sigma_p\right\}\right]
-
     .. [1] https://mr-easy.github.io/2020-04-16-kl-divergence-between-2-gaussian-distributions/
-
     """
     return 0.5 * (torch.log(torch.abs(q_var) / torch.abs(p_var)) - 1.0 + ((p_mean-q_mean)**2)/q_var + p_var/q_var)
 
@@ -47,6 +45,8 @@ def log_gaussian_kl(
     """Calculate KL Divergence of 2 Gaussian distributions.
 
     KL divergence between two univariate Gaussians, as derived in [1], with k=1 (dimensionality) and log variances.
+    .. math::
+        D_{KL}(p||q) = \frac{1}{2}\left[\log\frac{|\Sigma_q|}{|\Sigma_p|} - k + (\boldsymbol{\mu_p}-\boldsymbol{\mu_q})^T\Sigma_q^{-1}(\boldsymbol{\mu_p}-\boldsymbol{\mu_q}) + tr\left\{\Sigma_q^{-1}\Sigma_p\right\}\right]
 
     Parameters
     ----------
@@ -66,10 +66,6 @@ def log_gaussian_kl(
 
     References
     ----------
-    .. math::
-        D_{KL}(p||q) = \frac{1}{2}\left[\log\frac{|\Sigma_q|}{|\Sigma_p|} - k + (\boldsymbol{\mu_p}-\boldsymbol{\mu_q})^T\Sigma_q^{-1}(\boldsymbol{\mu_p}-\boldsymbol{\mu_q}) + tr\left\{\Sigma_q^{-1}\Sigma_p\right\}\right]
-
     .. [1] https://mr-easy.github.io/2020-04-16-kl-divergence-between-2-gaussian-distributions/
-
     """
     return 0.5 * (q_logvar - p_logvar - 1.0 + torch.exp(p_logvar - q_logvar) + ((p_mean - q_mean)**2)*torch.exp(-q_logvar))