GP KLD loss being negative

[amingc3]

Hi spaVAE team,
I tried running the model on a new dataset and encountered a situation where the GP KLD loss became negative. After debugging, I found that the initial GP KLD loss was positive, but after gradient descent, it turned negative. Is this behavior normal?

[ttgump]

Hi amingc3,
The GP KLD loss should not be negative. If possible, could you share the data with us so we can debug it? Thanks.