tomodrgn.convergence.calc_kld_two_gaussians

calc_kld_two_gaussians(z_mu_train: ndarray, z_logvar_train: ndarray, z_mu_test: ndarray, z_logvar_test: ndarray, workdir: str, epoch: int) → None

Compute the KLD between two gaussians with diagomal covariance. Used to test convergence via difference between each particle’s train and test embedding for selected epoch.

Parameters:

• z_mu_train – numpy array of shape (n_particles, zdim), latent embedding means deriving from particle train images

• z_logvar_train – numpy array of shape (n_particles, zdim), latent embedding log variance deriving from particle train images

• z_mu_test – numpy array of shape (n_particles, zdim), latent embedding means deriving from particle test images

• z_logvar_test – numpy array of shape (n_particles, zdim), latent embedding log variance deriving from particle test images

• workdir – str, absolute path to model workdir

• epoch – int, current epoch number

Returns:

None