A custom loss function for a VAE learning a multivariate normal distribution with a full covariance matrix
vae_loss_correlated(
encoder,
inv_skill_cov,
det_skill_cov,
skill_mean,
kl_weight,
rec_dim
)the encoder model of the VAE, used to obtain z_mean and z_log_cholesky from inputs
a constant tensor matrix of the inverse of the covariance matrix being learned
a constant tensor scalar representing the determinant of the covariance matrix being learned
a constant tensor vector representing the means of the latent skills being learned
weight for the KL divergence term
the number of nodes in the input/output of the VAE
returns a function whose parameters match keras loss format