cvignac / DiGress

code for the paper "DiGress: Discrete Denoising diffusion for graph generation"
MIT License
314 stars 68 forks source link

Normalization for congress #97

Open liun-online opened 2 months ago

liun-online commented 2 months ago

Hi Vignac,

Thanks for your code! I notice that ConGress normalizes the data for traning, and unnormalizes it for val_nll and sample. Besides, from continuous.yaml, only X is actually normalized. Could you help me for this reason, please?