Comparison between deep Auto-Encoder (stacked) and RBM training?

Hi.

im trying to implement a recommander system using RBM and deep auto-encoder architectures separately with pytorch. After the training i got a better performance result in RBM over AE.however,i have seen many papers and studies which said that the AE is always better than RBM .i used the same dataset for both architectures. and im trying to do a comparison between these two methods of deep learning.

My question is why did i got a different results than excpected? should i change something or it is ok ?