Dear author, regarding the compute_kl_divergence method in your utils.py, kl_div = means_term + trace_term-d + log_det_cov_p, your calculation here is different from the traditional kl divergence. What considerations were made? I noticed that in the log of the calculation, the kl divergence is negative from beginning to end. Shouldn't the kl divergence be non-negative?
Dear author, regarding the compute_kl_divergence method in your utils.py, kl_div = means_term + trace_term-d + log_det_cov_p, your calculation here is different from the traditional kl divergence. What considerations were made? I noticed that in the log of the calculation, the kl divergence is negative from beginning to end. Shouldn't the kl divergence be non-negative?