VAE evidence lower bound with closed-form Gaussian KL (hmelb)

Usage

morie_geron_elbo(
  x,
  mu,
  log_sigma,
  x_recon = NULL,
  likelihood = "gaussian",
  sigma_x = 1
)

Arguments

x: Inputs (m, d).
mu, log_sigma: Posterior params (m, k).
x_recon: Optional decoder output (default x).
likelihood: "gaussian"/"bernoulli".
sigma_x: Fixed decoder sd for the Gaussian likelihood.

Value

List with elbo, loss, kl, reconstruction_log_lik, per_sample_kl.

Examples

morie_geron_elbo(x = c(1, 2, 3, 4, 5, 6, 7, 8), mu = c(1, 2, 3, 4, 5, 6, 7, 8),
  log_sigma = c(1, 2, 3, 4, 5, 6, 7, 8))
#> $elbo
#> [1] -642317.6
#> 
#> $loss
#> [1] 642317.6
#> 
#> $kl
#> [1] 642316.7
#> 
#> $reconstruction_log_lik
#> [1] -0.9189385
#> 
#> $per_sample_kl
#> [1] 2.694528e+00 2.679908e+01 2.027144e+02 1.493979e+03 1.102023e+04
#> [6] 8.138890e+04 6.013191e+05 4.443079e+06
#> 
#> $per_sample_elbo
#> [1] -3.613467e+00 -2.771801e+01 -2.036333e+02 -1.494898e+03 -1.102115e+04
#> [6] -8.138981e+04 -6.013201e+05 -4.443080e+06
#> 
#> $latent_dim
#> [1] 1
#> 
#> $likelihood
#> [1] "gaussian"
#> 
#> $estimate
#> [1] -642317.6
#> 
#> $n
#> [1] 8
#> 
#> $method
#> [1] "ELBO = E_q[log p(x|z)] - KL(q||p) with closed-form Gaussian KL"
#>