search

PaddlePaddle / PaddleHelix

Bio-Computing Platform Featuring Large-Scale Representation Learning and Multi-Task Deep Learning “螺旋桨”生物计算工具集
Other
982 stars 221 forks source link

GEM pretrain model. #253

Open karthikjetty opened 1 year ago

karthikjetty commented 1 year ago

Hi! I was looking through the installation guide for ChemRL GEM at the following link. https://github.com/PaddlePaddle/PaddleHelix/tree/dev/apps/pretrained_compound/ChemRL/GEM.

I have downloaded the pretrained model trained on the ZINC dataset and I notice that it says "Also, the pretrained model can be used for other molecular property prediction tasks." I was wondering what tasks it would be effective at predicting, given that the demo ZINC dataset doesn't provide any task names for the model to be trained on.

For example, would this model be accurate at predicting any of the tasks mentioned in the other datasets that could be used for further pretraining of the model?

Noisyntrain commented 1 year ago

Hi karthikjetty, The pretrained model on ZINC can be further used to predict downstream datasets' tasks once added proper head(just like the way we use it in finetune_class.py and finetune_regr.py). Hope this can be helpful to you.