Norman tutorial question

[chujunhe]

Hello,

Thank you for sharing the tutorial notebook! I was trying to replicate the results in the Norman notebook: https://github.com/theislab/cpa/blob/main/docs/tutorials/Norman.ipynb.

I obtained consistent embedding space visualizations but pretty different numbers on the evaluation metrics. Could you please help advise if I missed anything? Thanks!

Results in the tutorial:

My results:

[tbheinen]

Hi, we are running into the same issue; it would be great to get an update on this.

There is at least one problem with the Notebook, it seems R2 scores are computed on raw counts / predictions rather than log-transformed and normalized inputs. I am not sure fixing this solves the OPs problem, however.