The problem of structure quantity generated and loss function.

[JHWang1001]

Dear txie,

I train the pervo data use

python cdvae/run.py data=perov expname=perov model.predict_property=True
python scripts/evaluate.py --model_path MODEL_PATH --tasks  opt

I have some question:

train_loss_step 7.70267
train_kld_loss_step 32.16307
train_kld_loss_epoch 33.9217

These numbers look like I didn't get a good convergence results. Does this mean that my training wasn't successful? Can you give me some examples for reference? Are they related to the number of structures generated? I only got 356 structures after train pervo. Sometime, I use my own data, I only can get 11 structures. XD. Can I get more structures with more training step?@txie-93

Best wishes and looking forward to your reply Jiahui Wang