Random seed about scarches

[M0hammadL]

@mxposed

ok, I added a feature in which you get the fixed z as I explained. You need to set
return_mean = True in network.get_latent(reference_adata,condition_key, return_mean=True)

please pull from repo to have this feature :)

[M0hammadL]

Hi Nik,

This comes up due to the underlying structure of get_latent function. When you project a cell to latent space you have a fixed value for that cell in z, however since the model is generative, a noise term will be added to this z. Thus, you have z = z_fix + N(0,1) in which N(0,1) is a sample from a random Gaussian that is why you get different values every time you use get_latent. There is a way to avoid this that I provide the ability to either get z or z_fix. However, in total there should not be a big difference that avoids reproducibility.