insilicomedicine / GENTRL

Generative Tensorial Reinforcement Learning (GENTRL) model
610 stars 218 forks source link

Valid rate of pretrained model #27

Open xiaohongniua opened 4 years ago

xiaohongniua commented 4 years ago

Thansk for your code! I completely trained the model with pretraining. However when sampling from the pretrained model , I found many smiles generated are invalid. Is it the reason that I did not run RL ?

ghost commented 4 years ago

Hi,I'm getting error while running pretrain code. If you're done with this kindly update me.

wjybill commented 2 years ago

The valid percentage for our pre-trained is about 20~30%, after reinforcement it will increase to 99% but more duplicated and narrow sampling results.