Reduction of the size of the latent vector

Hi!

I was having some trouble when transforming my dataset. 
The output lantent representation somehow shrinks the length of the tensor. 

Using the default settings.
before calling vrae.fit_transform(dataset)
the dataset has 3096  samples

now it has 3072
the shape is (3072, 10) # 10 is the latent_length

I wasnt able to find where the error occurs I hope you can help me.