minkaixu / confvae-icml21 Goto Github PK

Hi, I notice there are two kinds of generation functions in the eval file. And you used

def em_generate_batch(model: EdgeCNF, data, num_samples, embedder=Embed3D(), dg_init_pos=None, em_steps=0):

to produce molecular conformations. However, this process is different from what the Imodel is trained when calculating the reconstruction loss. I also notice that you set the em_step to 0.

Did you actually use this method for the generation? If so, is this more sophisticated approach better than the simple one? Can you provide more details about this method, or perhaps give me some references?

Great thanks!

Hi！thank u for sharing ur work
How long does this default model take to train?
Can you provide a pre-trained model?

Hi, thanks for sharing the code.

However, when I reran the code without any modification. I came across the following issue, can you please give any guidance or help to solve it?

Hello

Thank you for sharing this job for fast conformer generation.

And I hope to try this command "python eval_vae.py
--ckpt ./logs/VAE_QM9
--dataset ./data/iclr/qm9/test_QM9.pkl
--num_samples -2" to see how it works using trained model to generate conformers and cannot find where the "./logs/VAE_QM9
" file is.

I was just wondering would you mind sharing the .ckpt file for testing.

Thank you.