cvignac / DiGress

code for the paper "DiGress: Discrete Denoising diffusion for graph generation"
MIT License
354 stars 71 forks source link

Invalid Molecule Obtained #19

Closed anar-rzayev closed 1 year ago

anar-rzayev commented 1 year ago

Hi Clement, thank you for the constant updates these days in the repository. I wanted to ask you about one strange error I get while running the script python3 main.py +experiment=debug.yaml

I have installed all the necessary packages, so far following your instructions on environment installation. However, before launching the full experiments, I wanted to run the debugging code and see how it performs.

Снимок экрана 2023-03-17 в 17 17 37

Unfortunately, having outputted the ret = run_job() and 3 successful 100 % processing stages, it starts to give "Invalid molecule obtained" while Converting the QM9 dataset to SMILES for remove_h=True...

Do you know the reasons why this may occur? Thanks for the help!

cvignac commented 1 year ago

This is normal, some molecules from QM9 cannot be parsed to SMILES. One reason is that DiGress does not handle formal charges, but it is not the only reason. In the paper (Table 2), you can see that the validity is 99.3% for the dataset itself, which means that you should get 0.7% of "Invalid molecule obtained".

Clement

anar-rzayev commented 1 year ago

Oh true, it makes sense. Thank you Clement for the references. I am closing the issue!