search

ehoogeboom / e3_diffusion_for_molecules

MIT License
432 stars 113 forks source link

Training Split for G-SchNet #28

Closed ameya98 closed 10 months ago

ameya98 commented 1 year ago

Hi, was the G-SchNet model used to generate samples here trained on the same data-split? The original paper only used 50k training samples.

pearl-rabbit commented 1 year ago

Hi, I also have a question about the G-Schnet model. I sampled 10,000 molecules using the pre-trained model given by G-Schnet and evaluated it using the metrics of this paper, and the results for validity and uniqueness are >95%. Do you know the reason for this, please?

ehoogeboom commented 10 months ago

@ameya98 we used the one as defined in the paper, that might be different than g-schnet. @pearl-rabbit the samples from gschnet we used (and the corresponding python analyze file) are both in this repo