omicsml / scdiff Goto Github PK

I find that there are three test metrics relative to correlation of imputation: denoise_corr_normed│ denoise_corr_normed_all │ denoise_global_corr_normed.
I wonder what are their differences? And what's the meaning of 'global'? Thanks a lot!

Hi, thanks for your great work! I wonder if we can have an API to call and train the model like what GEARS did. Thanks a lot.

Can you provide trained models? Can you also provide examples/notebooks on how to use a model for various tasks such as imputation/classification/sampling?
Thanks

Hi,

May I know how the authors process the data? For example, how do you normalize the data?

Thank you very much for your help.

Hi,

I found that the x update does not rely on the model itself during denoising except for the last time step. In the function p_sample_loop, the update of x is covered by the function self.prepare_noised_input, which is shown below:

Could you please explain it since it seriously affects the results of the paper?

Thank you very much!