search

cvignac / DiGress

code for the paper "DiGress: Discrete Denoising diffusion for graph generation"
MIT License
349 stars 73 forks source link

Questions on the exact setup used for the experiments #58

Open Asduffo opened 1 year ago

Asduffo commented 1 year ago

Hello Mr. Vignac

I am trying to reproduce the results you have obtained on the experiments which focused on the diffusion guidance, and I would like to do so with the exact same setup you have used for such experiments (meaning without the various bug fixes). It is kinda difficult to navigate through the various commits, but it seems to me that the commit at https://github.com/cvignac/DiGress/commit/ea23692881c875ba74a410fad7d5dcad6748dbf9 is the one you have used for such experiments. Am I correct? May I also ask you whether the hyperparameters you used to train the main model were the ones listed in configs/experiment/qm9_no_h.yaml?

Many thanks in advance. Regards.

pricexu commented 8 months ago

I have a similar question. If it is possible to provide some straightforward command to reproduce the results? E.g., I tried the command: python3 main.py dataset=planar +batch_size=8 but after training for several epochs, it keeps showing "Visualizing molecules... Done. Visualizing chains...".

How can I skip those steps? Personally, I am not familiar with the packages hydra and its config system, so some most direct commands to present the results in the tables from the paper would be greatly appreciate!

Bests